It really feels like something affected people's ability to just deal with minor inconveniences. I thought it was social media mass psychosis, but then I went out today and people were freaking out that the power and internet was going to go out at any minute.
Marcus Hutchins
Marcus Hutchins (born 1994 in Ascot), also known online as MalwareTech, is a British computer security researcher known for stopping the WannaCry ransomware attack. As of 2025, he is a Principal Threat Researcher at managed detection and response firm Expel. More on Wikipedia
-
Marcus Hutchins
@MalwareTechBlog
-
Marcus Hutchins reposted
himug-lamuh
@HimugLamuh
the supreme leader has shown remarkable forgiveness by letting @MalwareTechBlog do a closing keynote
The North Korean Computer Emergency Response Team
@dprkcert
#bsidespyongyang2026 | @BSidesPyongyang
x.com/i/broadcasts/2YRZpqaqJPgrak -
Marcus Hutchins
@MalwareTechBlog
It's wild how much money people are pouring into a completely economically unviable company that has like a year of runway left at best.
unusual_whales
@unusual_whales
BREAKING: Anthropic's IPO has drawn nearly $500 billion in investor orders, per CNBC.
Institutional orders reached about $400 billion after the books closed a day early, while retail orders are approaching $100 billion. That's roughly seven times the shares on offer, with multiple single orders of $10 billion or more. Demand reportedly topped SpaceX's IPO.
Anthropic is now considering selling more shares and pricing above $80, up from its original $70–78 range. The original deal was 825 million shares, raising up to $64.4 billion. -
Marcus Hutchins
@MalwareTechBlog
If you hear anyone claiming "AI is enabling unskilled criminals/hackers to operate with the sophistication of elite state-sponsored hackers", you can safely unfollow them and ignore all their future takes on cybersecurity.
As someone who monitors threat actors professionally, and has watches thousands of unskilled hackers try to figure out how to hack with AI, it's basically just this:
Sometimes they succeed, a lot of the time they don't. But even when they do, it doesn't come within 1000 miles of anything resembling a sophisticated state-sponsored intrusion. -
Marcus Hutchins
@MalwareTechBlog
As much as this suck, it's a pretty good time for AI to have happened. Many devices have layers on layers of exploit mitigations, which makes exploiting vulnerabilities significantly more complex, and often requires chaining multiple vulnerabilities together, and/or version specific exploits.
If this had happened in the pre ASLR + DEP era, the entire internet would basically be hacked. Instead, it's mostly just a lot of patching theoretical vulnerabilities, with the occasional exploitation. -
Marcus Hutchins
@MalwareTechBlog
Adding "sponsored by GLM 5.3" to the top of my website because US frontier labs can eat shit and die.
MTS
@MTSlive
SITUATION DETECTED: The White House and Treasury are discussing a proposal that would require US companies to publicly disclose the use of Chinese open-weight AI models on their websites, aiming to discourage their use through public pressure, per Reuters.
-
Marcus Hutchins
@MalwareTechBlog
I'm Keynoting BSides Pyongyang at 3 PM ET, come check it out! Conference runs from 10:30 AM ET until 4 PM ET, tickets are free and all talks are livestreamed.. Lots of good content worth watching :)
bsidespyongyang.com/2026/ -
Marcus Hutchins
@MalwareTechBlog
Another simple explanation is that an average software update has a low to medium chance of breaking the system. Previously, orgs navigated this by performing extensive testing, then rolling out updates a few systems at a time.
With the severity of some of these security issues being fixed, they can't afford to wait that long. The options are: risk breaking something by rolling out updates too fast, or get hacked and risk lots of things getting broken.
Given that we just had a major vulnerability in PHP, and many organizations run PHP, it's not unexpected that we would see some major companies having software outages. PHP applications are highly version specific. Attempting to update PHP is a very well known cause of software breaking.
Now, combine that with the fact that many orgs are running with less than the minimum viable number of employees. They got overexcited about AI and tripped over their own feet rushing to preemptively lay off double digit percentages of their workforce.
I'm not surprised that there's a ridiculous number of major outages. I'm surprised that there aren't far more. It's amazing some of these companies have managed to hold it together as long as they have. But expect it to only get worse from here on. -
Marcus Hutchins
@MalwareTechBlog
I shitposted so hard about the outages that now I'm getting invites to go on major news networks and talk about the current string of outage. Unfortunately, all the invites are for morning segments on east coast time, and I'm not committed to the bit enough to wake up at 4am
-
Marcus Hutchins
@MalwareTechBlog
Can the FBI issue a flash warning that Marcus is really bored of hearing about Artex and would like to spend his time looking at literally anything else.
-
Marcus Hutchins
@MalwareTechBlog
Its fun watching people try to form conspiracies. "You really think all these outages are a coincidence? The government is hiding something!". Yeah, they're hiding the fact that every tech company replaced like 80% of their employees with bots that guess words, hardly anyone left knowns wtf they're doing, and admitting it's not working out would implode the economy because people invested half the US GDP into this shit.
-
Marcus Hutchins
@MalwareTechBlog
You wouldn't believe the number of networks I've seen saved by canary credentials, this stuff works and it works extremely well.
The first thing most threat actors do when they successfully break into a network is look for useful credentials to pivot off of or escalate access with. Make sure they find some.
fake API keys, usernames and passwords, session tokens, NTLM hashes, password.txt. Put them everywhere. Servers, employee endpoints, password managers, the web browser's password/session databases, cloud storage buckets, network shares.
The credentials should look legitimate, high value, and be unique to each individual system or user. That way you can set up granular detection for what happens when an attacker attempts to log in with a canary credential.
You can not only instantly lock out the system it was stolen from, but also any real credentials belonging to that system/users, which may have been exfiltrated in the process. -
Marcus Hutchins
@MalwareTechBlog
So does the US have like a secret ATC backup system, or can adversaries just target the same single point of failure and ground all US air traffic?
I keep being told the US is a super power, but all I'm seeing is bumbling middle power shit. Wym the emails and planes are down? -
Marcus Hutchins
@MalwareTechBlog
I meant in that specific instance.
Google's security teams are full of great people who genuinely want to make a difference, any I just sometimes disagree with the decisions they make.
Google as a company, though, is a neon warning sign pointing out the perils of late-stage capitalist societal decay brought on by a political system that puts shareholder interests above all else, in a country currently collapsing under the weight of excessive deregulation.
My personal experience with Google so far has been: their old 'featured snippet' search engine feature kept telling people I was behind the largest most damaging cyberattack in history (which I actually stopped, not created). Then when I asked them to fix it, the head of search acted like I was being a massive inconvenience for asking them not to claim I ransomed hospitals and destroyed hundreds of thousands of systems across 150 countries.
Then when they rolled out AI search, they vaporized 80% of my website traffic by having their LLM plagiarize my articles and present a slightly-reworded version to users, claiming it was "transformative" and therefore not copyright infringement. This was part of an intentional scheme to increase ad revenue by plagiarizing every website's content, keeping users on Google where they'd see Google's ads, instead of directing them to the source website (which was like the entire purpose of Google).
Trust me, there's no love lost between me and Google, or between me and this techno-feudalist hellscape I find myself trapped it. Google is easily one of my top 5 least favorite companies on the planet, but their security team does good work.
I will remain objective on issues that require it, and save my Google bashing for other times.
Mild Billy Donovan
@MildBillDonovan
"I don't think Google has any malicious intent at all"
Ok that's where you're completely wrong.
Marcus Hutchins
@MalwareTechBlog
FFMPEG had one of the most valid crash outs in infosec history IMO. There's a huge elephant in the room when it comes to the fact that cybersecurity is extremely unbalanced. You have like 1,000 people pointing at problems for every 1 person fixing them.
Vulnerability research is extremely important, but it's become somewhat of a marketing stunt. First with fuzzers, and now with LLMs. It's a lot of "look what my new toy can do" rather than earnest attempts at actually fixing security problem.
What Google did to open source developers was especially egregious. They're a $3.6 trillion dollar company, who spent billions upon billions of dollars building a bot that points at problems really really fast, then they just dumped them all on a bunch of unpaid open source project maintainers to fix.
Then because of Google's 90 day disclosure policy, the maintainers have 90 days to fix the issues. After 90 days, Google goes public with the issues, making their project look bad if the vulnerability is yet to be fixed, and also potentially harming users if it gets exploited.
The 90 day policy was intended to give developers a reasonable amount of time to fix issues, while also applying some pressure with the threat of disclosure. Otherwise, you do get a bunch of goober ass vendors that will just ignore all security issues reported to them, then act completely surprised when their software gets hacked.
But when you're a small team, non-profit, or a solo developer, and you have a trillion dollar company's AI just blasting you with infinite problems to fix (each coming with a 90 countdown & a threat), that becomes extremely untenable. They're essentially inadvertently blackmailing developers into becoming unpaid interns.
Then on top of that, Google profits unimaginable amounts of money from the use of open source code. They use FFMPEG's code extensively across several highly-profitable products, meanwhile FFMPEG's developers work for free. There's no reason Google couldn't have donated money, engineers, AI tokens, or something towards fixing the problems. They can more than afford it.
I don't think Google has any malicious intent at all, it's just that the whole incentive structure behind vulnerability research is completely whack. All the glory goes to people who find issues, not those who fix them, so there's not a whole lot of interest in helping with fixes. Some vulnerability researchers will submit fixes with their reports, but most don't.
Honestly, in 2026 cybersecurity basically just feels like the equivalent of 50,000 people following the housecleaner around yelling "you missed a spot". Abysmally little effort has gone into anything outside of pointing out vulnerabilities.
Shirley Malik
@ShirleyYerJokin
FFMPEG was right all along and people gave them too much shit for it
Marcus Hutchins
@MalwareTechBlog
I get what Google is trying to do, but I don't agree with it at all. Sure, there are some vendors who just sit on bugs forever and don't make any effort to patch them. But there's way more overworked & underpaid teams, non-profits, and solo projects who are patching bugs as fast as they physically can.
The number of CVEs found in 2026 is double that of 2025, so double the workload for anyone doing security fixes. Then on top of that, reducing the disclosure time from 90 to 60 days gives them 2/3 of the time to do double the work.
At some point the dam is going to break. You have vulnerability researchers and AI companies burying software maintainers under an endless torrent of bug reports, while doing little to nothing to help with actually fixing the bugs. Eventually, the software maintainers are just going to burn out and give up.
Natalie Silvanovich
@natashenka
In the age of AI, 90 days isn’t fast enough. Today, Project Zero updated our disclosure policy to reflect the changes needed to protect users in this new world.
projectzero.google/2026/11/disclosure-policy.html -
Marcus Hutchins
@MalwareTechBlog
Definitely not the worst way the FBI has my weekend, but blasting out a warning about Artex at 3pm on a Friday was certainly a choice.
As far as I can tell, nothing new, nothing changed, activity is a month old. But it scared the customers, so now I'm spending my weekend "looking for Artex" 🥲 -
Marcus Hutchins
@MalwareTechBlog
I'd go as far as to argue that a lot of professional software engineers couldn't build an infostealer that comes close to the ones you can rent for ~$150/month (even with the help of AI). Operating costs would likely be higher too, since in-house dev loses to economies of scale.
Daniel Kelley
@danielmakelley
I keep seeing the claim that AI is what made cybercrime accessible to people with no skills. I get why, but you could already rent malware, hosting, phishing kits and customer support 10 years ago, and honestly none of it needed you to write a single line of code.
-
Marcus Hutchins
@MalwareTechBlog
The other thing that kind of bugs me (pun intended), is the "31% of intrusions are the result of software exploitation" statistic.
It hides two different problems:
Zero-day exploitation - the bad guys find a vulnerability before the good guys, then use it to hack into systems (a problem finding and reporting vulnerabilities reduces).
N-Day exploitation - someone reports a vulnerability, a fix gets published, but then the threat actors rush to use the information from the fix to build an exploit before downstream users get a chance to install the fix (a problem finding and reporting vulnerabilities exacerbates).
Since organizations typically take a long time to install security fixes, N-day exploits are extremely powerful, and also make up the majority of all software exploitation. There's also 100x more people who can create N-days from known vulnerabilities than can find and exploit Zero-days.
The more vulnerabilities found, the more vulnerabilities N-day attackers have to choose from, and the more software fixes that downstream users have to juggle.
Orgs struggle to even install the monthly Windows update within a month of it going live. Now mix in "every single product you use has critical vulnerability fixes every patch cycle so you have to install all 4,000 patches every month". And yeah.... It's pretty bleak.
That's not to say vulnerability research shouldn't be done, isn't important, or isn't a net long term benefit (I feel like I have to say this because half the people on this fucking website have zero capacity for nuance and try to read every post in the most disingenuous way possible).
But since the majority of exploitation is N-day exploitation, and vulnerability research increases N-day exploitation, using exploitation rate as a justification for pouring more resources into find vulnerabilities (but not helping fix or patch them), is just circular reasoning.
Plenty of good arguments for why vulnerability research is important, but "we're showing the attackers more vulnerabilities that they can use, which is leading to them using more vulnerabilities" is not one of them.
Equal effort needs to go into all part of the vulnerability management pipeline for this to work. But currently the effort drops by a factor of 10 from finding bugs to fixing bugs, then another factor of 10 from fixing bugs to ensuring the fixes get installed.
Just reducing the disclosure window and telling everyone "don't worry, I'm sure you'll figure it out" is extremely likely to make the problem worse, not better. -
Marcus Hutchins
@MalwareTechBlog
FFMPEG had one of the most valid crash outs in infosec history IMO. There's a huge elephant in the room when it comes to the fact that cybersecurity is extremely unbalanced. You have like 1,000 people pointing at problems for every 1 person fixing them.
Vulnerability research is extremely important, but it's become somewhat of a marketing stunt. First with fuzzers, and now with LLMs. It's a lot of "look what my new toy can do" rather than earnest attempts at actually fixing security problem.
What Google did to open source developers was especially egregious. They're a $3.6 trillion dollar company, who spent billions upon billions of dollars building a bot that points at problems really really fast, then they just dumped them all on a bunch of unpaid open source project maintainers to fix.
Then because of Google's 90 day disclosure policy, the maintainers have 90 days to fix the issues. After 90 days, Google goes public with the issues, making their project look bad if the vulnerability is yet to be fixed, and also potentially harming users if it gets exploited.
The 90 day policy was intended to give developers a reasonable amount of time to fix issues, while also applying some pressure with the threat of disclosure. Otherwise, you do get a bunch of goober ass vendors that will just ignore all security issues reported to them, then act completely surprised when their software gets hacked.
But when you're a small team, non-profit, or a solo developer, and you have a trillion dollar company's AI just blasting you with infinite problems to fix (each coming with a 90 countdown & a threat), that becomes extremely untenable. They're essentially inadvertently blackmailing developers into becoming unpaid interns.
Then on top of that, Google profits unimaginable amounts of money from the use of open source code. They use FFMPEG's code extensively across several highly-profitable products, meanwhile FFMPEG's developers work for free. There's no reason Google couldn't have donated money, engineers, AI tokens, or something towards fixing the problems. They can more than afford it.
I don't think Google has any malicious intent at all, it's just that the whole incentive structure behind vulnerability research is completely whack. All the glory goes to people who find issues, not those who fix them, so there's not a whole lot of interest in helping with fixes. Some vulnerability researchers will submit fixes with their reports, but most don't.
Honestly, in 2026 cybersecurity basically just feels like the equivalent of 50,000 people following the housecleaner around yelling "you missed a spot". Abysmally little effort has gone into anything outside of pointing out vulnerabilities.
Shirley Malik
@ShirleyYerJokin
FFMPEG was right all along and people gave them too much shit for it
Marcus Hutchins
@MalwareTechBlog
I get what Google is trying to do, but I don't agree with it at all. Sure, there are some vendors who just sit on bugs forever and don't make any effort to patch them. But there's way more overworked & underpaid teams, non-profits, and solo projects who are patching bugs as fast as they physically can.
The number of CVEs found in 2026 is double that of 2025, so double the workload for anyone doing security fixes. Then on top of that, reducing the disclosure time from 90 to 60 days gives them 2/3 of the time to do double the work.
At some point the dam is going to break. You have vulnerability researchers and AI companies burying software maintainers under an endless torrent of bug reports, while doing little to nothing to help with actually fixing the bugs. Eventually, the software maintainers are just going to burn out and give up.
Natalie Silvanovich
@natashenka
In the age of AI, 90 days isn’t fast enough. Today, Project Zero updated our disclosure policy to reflect the changes needed to protect users in this new world.
projectzero.google/2026/11/disclosure-policy.html -
Marcus Hutchins
@MalwareTechBlog
Just a head's up, it looks like LearnWorlds may have been hacked, and their mail server is currently being used to phish YouTube accounts (LearnWorlds is a major white-label online course platform, so you may not realize you're a user of their platform).
This morning I received an email from LearnWorld's official mail server, claiming something I posted on my YouTube channel violated the company's copyright, and that I need to respond immediately.
The email felt off. It cited no specific copyright violations, didn't reference my channel by name or link to any videos, was sent on a Saturday morning, and YouTube has a built-in copyright system to handle exactly this.
On the second pass I realized the link it asks recipients to visit to handle the "copyright claim" is a YouTube redirect link. It uses the official YouTube url, but has an embedded redirect which takes you to a third-party website.
The site is designed to look like an official YouTube webpage that enables users to review copyright claims against their channel. It contains both the YouTube and Google branding, as well as a very similar theme to the official YouTube one.
The site asks users to look up their channel name to check for copyright claims, presents them with a fake copyright violation, then asks them to "verify" their channel.
The channel verification button takes the user to an official Google OAuth sign in page, which if they log into, will give the hackers control over the Google account they sign in with, enabling them to steal both the Google account and associated YouTube channel. -
Marcus Hutchins
@MalwareTechBlog
I get what Google is trying to do, but I don't agree with it at all. Sure, there are some vendors who just sit on bugs forever and don't make any effort to patch them. But there's way more overworked & underpaid teams, non-profits, and solo projects who are patching bugs as fast as they physically can.
The number of CVEs found in 2026 is double that of 2025, so double the workload for anyone doing security fixes. Then on top of that, reducing the disclosure time from 90 to 60 days gives them 2/3 of the time to do double the work.
At some point the dam is going to break. You have vulnerability researchers and AI companies burying software maintainers under an endless torrent of bug reports, while doing little to nothing to help with actually fixing the bugs. Eventually, the software maintainers are just going to burn out and give up.
Natalie Silvanovich
@natashenka
In the age of AI, 90 days isn’t fast enough. Today, Project Zero updated our disclosure policy to reflect the changes needed to protect users in this new world.
projectzero.google/2026/11/disclosure-policy.html -
Marcus Hutchins
@MalwareTechBlog
It's fascinating how AI has just completely erased people's memory of pre-ChatGPT era automations. Not only were most hacking techniques already automated before "AI", many of those automations were actually faster and more effective.
-
Marcus Hutchins
@MalwareTechBlog
One of the startups I'm a founding investor of actually wanted to pivot into OT/ICS security and I was just like Nonononononono.
They were like "but it's a massive untapped market"
And I'm like "yeah, in the same way that erupting volcanos are untapped heated swimming pools". -
Marcus Hutchins
@MalwareTechBlog
The way the media reports on tech outages is dumb as hell. Any given major tech outage is about as likely to be terrorists or squirrels, as it is to be a cyberattack. Yet every time "no indication yet of a cyberattack". No indication yet if Bin Laden is back from the dead either.
Marcus Hutchins
@MalwareTechBlog
Every time there's any tech outage I get like 50 people messaging me asking if it's a cyberattack, and the only reason I haven't set up an auto-responder that just responds to "is it a cyberattack" with "no" is maybe 1 time it actually will be, then I'll look stupid.
-
Marcus Hutchins
@MalwareTechBlog
What we thinking on the nationwide ground stop?
-
Marcus Hutchins
@MalwareTechBlog
An open secret in cybersecurity is that this is how insecure a lot of ICS/OT infrastructure is, including a decent amount of critical infrastructure like water & power distribution systems.
This stuff never needed to be secure because nobody bothered hacking it. Financially motivated hackers were busy ransoming fortune 500 companies, and critical infrastructure attacks during peacetime would be extremely escalatory, so hostile state actors left it alone.
But now you have multiple active military conflicts involving nation that have major cyber capabilities, so the technical debt is coming due all at once. Russia is engaging in sabotage operations across Europe, and Iran is throwing hands at anyone who aligns against them.
The good news is, most of these attacks are very haphazard and therefore not particularly impactful. Iranian attacks on US water infrastructure has led to minimal disruption, and the fridge attacks caused some fairly significant damage, but the threat to life was low. However, all of this is just a warning of what's to come.
I personally see it as a bit like cybersecurity equivalent of how some vaccines work. The vaccines introduce a weakened version of the bad thing into your body to teach the immune system how to fight it, protecting you from a much worse thing that can kill you.
These attacks, if coordinated and well planned out as part of a direct military confrontation, could kill people. So it's better that providers are finding out that their security absolutely sucks right now, instead of during the next major war. It gives them (and the government), plenty of time to plan and prepare.
Just based on how much infrastructure China has infiltrated in the past 5-10 years, I suspect a direct confrontation with them would be absolutely catastrophic if it were to happen right now. They've been making very serious and recurring efforts to hack into and embed themselves inside all kinds of US infrastructure, including critical infrastructure.
Many analysts believe that China's strategy for Taiwan would involve using cyberattacks to impose massive cost on the US if it were to attempt to come to Taiwan's aid. Disruption of military and military-adjacent infrastructure to slow a US response is extremely likely.
But there's also the much more extreme strategic option of just causing massive disruption to daily life. The US public have repeatedly show that they have an extremely low tolerance for wars that have domestic consequences, and that's a really easy pressure point that US adversaries love to use.
Iran manipulating gas prices was enough to help swing federal, state, and local elections. Support for arming Ukraine quickly evaporated over empty threats made by Russian, and propaganda convincing the public that the war was taking away money that would have otherwise gone to them.
Iran tends to just break into exposed critical infrastructure controllers and mess with the setting to try and break stuff, but someone onsite is usually able to mitigate any impact relatively quickly. Both Russia and China have much more sophisticated capability to cause extended disruption which takes disaster recovery plans into account.
Russia has actually demonstrated such capabilities publicly in the past. Back in 2015 they took out 30 something power substations across Ukraine, causing power outages that spanned 3 provinces, including the capital. They then did it again in 2016.
I suspect an attack on the US would go much, much, worse. Ukraine was far more prepared for a direct confrontation with Russia, whereas the US kinda just lets private companies run most of their critical infrastructure however they see fit, which usually means a lot of cutting corners to increase profits.
Additionally, the US let a 50 something year old billionaire man-child decimate the agency responsible for critical infrastructure security, along with other parts of the US' national security apparatus. Turns out, it all costs the government money, which could have been put to better use propping up an overvalued car company.
There's rather a lot of work to do, and when I say a lot, that is a comically large understatement. But fortunately, there is some time left to do it.
Marcus Hutchins
@MalwareTechBlog
I love spicy vulnerability research blogs. They pretty much verbatim said that the fridge controller code was so bad that they decided it would be too much effort to set up a harness, instead audited the code by hand, at which point they found 6 critical pre-auth RCEs in under an hour.
www.midnightblue.nl/blog/multiple-rce-danfoss-aksm800a -
Marcus Hutchins
@MalwareTechBlog
I love spicy vulnerability research blogs. They pretty much verbatim said that the fridge controller code was so bad that they decided it would be too much effort to set up a harness, instead audited the code by hand, at which point they found 6 critical pre-auth RCEs in under an hour.
www.midnightblue.nl/blog/multiple-rce-danfoss-aksm800a
Marcus Hutchins
@MalwareTechBlog
I guess this was what Mr Abliterated GLM 5.3 was talking about. He was able to get the LLM to find bugs so obvious that someone found them by hand with the CTRL + F key lol.
-
Marcus Hutchins
@MalwareTechBlog
Super hyped for Microsoft’s next product. Now that they’ve 10x’d every engineer with AI and stopped wasting resources on pointless stuff like keeping the infrastructure that runs 90% of the planet online, I have high expectations. I’m thinking time machine or resurrecting Jesus.
-
Marcus Hutchins
@MalwareTechBlog
It’s clear that Microsoft is struggling as a result of not replacing enough employees with AI. Look at this grammatical error. They couldn’t even make it to 20 hours of nonsensical corporate slop posts about their outage without a grammatical error. Copilot would never 💔
Microsoft 365 Status
@MSFT365Status
We're engaged in multiple workstreams to alleviate impact, address our current blocker, and returning portions of service infrastructure back to a healthy state. A more comprehensive update can be found at MO1550821 in the Microsoft 365 admin center, as well as status.cloud.microsoft if the admin center is not available to you.
Marcus Hutchins
@MalwareTechBlog
How am I supposed to not understand what’s going on with their 17h outage when there’s grammatical errors? They ruined perfectly good meaningless AI slop that says nothing at all. Extremely disappointing. What next? Reviewing the AI’s broken code before it gets pushed to prod?
-
Marcus Hutchins
@MalwareTechBlog
Wait, are you suggesting that telling people your product is going to take their job and then murder them wasn't a winning marketing strategy?
roon
@tszzl
the chinese have rightfully earned enormous global goodwill and soft power with their open source model efforts. even most Americans in this space feel more sympathetic to them than with the model companies
-
Marcus Hutchins
@MalwareTechBlog
True, you got me.
The UK doesn't classify documents at all, they just store them at the local public library. They prevent leaks by having some guy named Reginald Bartholomew the 3rd who comes to your house and calls you a twat if you send them to the wrong person.
Magdovus
@Magdovus
Given that the UK doesn't use secret or top secret classification, I'm going to assume you're full of shit
-
Marcus Hutchins
@MalwareTechBlog
I can't figure out if the spotlight pushed Jensen off the deep end, or if he was always like that. AI demand made his company valuation skyrocket, then suddenly everyone is shoving cameras in his face and treating him like Jesus. Some of the stuff he says in interviews is like actually insane.
-
Marcus Hutchins
@MalwareTechBlog
One of my agents suggested quitting my job to focus on content creation full time, because they think they can bill more than my day job. I was thinking either this dude is about to quiet quit, or vastly overestimates my income, because they're already billing like 4x that.
Post not available here, view on X
Marcus Hutchins
@MalwareTechBlog
But when you work at a security services provider, basically every day is a 5 alarm fire cybersecurity incident somewhere. So it's hard to figure out how to not end up in permanent burnout mode because you day job is stressful af, and you run a whole business on top of that.
Marcus Hutchins
@MalwareTechBlog
I definitely should have picked a different niche. Being a travel influencer or doing OnlyFans seems like it'd be way less stressful than being waterboarded with cortisol on a daily basis 😆
-
Marcus Hutchins
@MalwareTechBlog
I own around $20,000 in NVIDIA Put options, so either I'm going to wake up tomorrow and have the best day ever, or the market is going to do something extremely irrational and I'm getting another lube-free fisting.
Marcus Hutchins
@MalwareTechBlog
98% profit lmao
-
Marcus Hutchins
@MalwareTechBlog
I bought NVIDA puts because I thought the AI bubble was going to pop, but instead a second mistress has hit the tower
The Spectator Index
@spectatorindex
BREAKING: Nvidia is seeking to find a new CEO to replace Jensen Huang, according to Bloomberg report.
-
Marcus Hutchins
@MalwareTechBlog
Maybe the US version is different, but in the UK, secret clearance is so useless it was less paperwork than the background check for my corporate job.
Top Secret they made me list out the full legal name of every single person I've ever fucked and then went and interviewed them.
Honeypotting someone and finding out they only have secret clearance gotta be embarrassing af. Just going back to their handler like "I figured out what vehicles they use, it's something called a Humvee".
Knowing Better
@KnowingBetterYT
When I was in the Army, I got a secret-level security clearance. Not once did a foreign honeypot try to message me, which is crazy given what I know.
Still know, really. It's not like you forget that stuff. They made us memorize it. Anyway, sometimes I talk in my sleep. -
Marcus Hutchins
@MalwareTechBlog
Me explaining to AI dudes that cybersecurity ≠ vulnerability management and that the majority of cyberattacks involve exactly zero vulnerabilities
Jeremiah Grossman
@jeremiahg
Project Glasswing:
* 37,966 findings.
* 2 exploited in the wild.
www.vulncheck.com/blog/anthropic-glasswing-november-update -
Marcus Hutchins
@MalwareTechBlog
At what point does an IPO just become a cap raise that bypasses Rule 506? Nothing about what Anthropic is doing is an IPO. They just can't raise any more money from private institutions so they're going public to try and raise from retail investors.
Marcus Hutchins
@MalwareTechBlog
I really had to submit financial statements proving I'm an accredited investor to invest in farmland because it's considered "risky", but apparently yeeting all your money into a GPU rental company meme stock run by a doomsday cult is fine 🙃
-
Marcus Hutchins
@MalwareTechBlog
Surprised by France attributing the fridge attack to Iran. My guess would have been one of the Russian state-sponsored sabotage groups that targets random EU infrastructure. I don't think France played any role in the Iran war, but tbh Iran has been throwing hands at everybody.
Marcus Hutchins
@MalwareTechBlog
According to a comment on my LinkedIn post, France was involved with shooting down a decent number of Iranian drones over the Middle East, which would 100% explain it. Iran basically considers anyone who provided even defensive assistance to be an aggressor, and therefore a legitimate military target.
What it doesn't explain is the much larger second wave of fridge attacks, which occurred last week, and hit around 9 EU nations. Several of those nations don't appear to have had any role at all in the Iranian war at all, but would be targets of interest for Russia.
Entirely speculation, but Russia has in the past repurposed other nation's cyberattacks for their own purposes. So I would not be surprised if it turns out the first wave of attacks was Iran and second wave was Russia, but also entirely plausible both were Iran and the reasoning is yet to become clear. -
Marcus Hutchins
@MalwareTechBlog
Ram Altman is gonna hunt me for sport when he finds out every PC I own has a minimum of 128 GB of DDR4 and several are DDR5.
-
Marcus Hutchins
@MalwareTechBlog
Imagine if a Huangbang is what ends up popping the AI bubble
-
Marcus Hutchins
@MalwareTechBlog
I think the invasion of Ukraine is a really solid case study on what an actual "cyber war" looks like. Russia pulled off a couple of massive cyber attacks early on in the war (and before the war), but once the shooting started, they focused almost all their efforts towards kinetic warfare.
I suspect this is due to the fact that cyber warfare capacity is a subset of intelligence capacity, and Russia quickly realized that their intelligence capabilities are better put to use aiding (much more impactful) kinetic attack, or helping defend against enemy kinetic attacks, rather than annoying civilians with random outages.
I'm also moderately certain this is how it'd go in any war. An initial burst of cyberattacks as powers deploy capabilities their nations had built out pre-war, but then a strong pivot to focusing on support for kinetic attacks as things run their course.
Unlikely kinetic attacks, most cyber attacks aren't highly repeatable and reusable. Taking out power substations requires a couple of missiles, or years of work building, planning, and testing cyber capabilities, which have components that are often specific to individual software/hardware brands. Once the adversary figured out the what & how, it's often possible for them to mitigate the effects and/or prevent a repeat attack.
Meanwhile, every nation can mass produce kinetic capabilities, and the "just send more missile" strategy tends to work much better than even the most sophisticated cyberattacks possible.
Similarly, for a pure cyberwar, a cyber power would just have to, for some reason, wish to tie their own hands behind their back by taking all kinetic capabilities off the table. Given how much cyberattacks tend to impact civilians, I don't see any reason why any nation would commit to this.
The idea also runs into huge issue when it comes to conflicts involving less technologically reliant nation, especially the ones that can and do turn of the entire internet. They can cause way more harm to you than you can to them.
Cyber attacks are really just not that effective relative to the time and cost of producing them, and "cyberwar" being the future of warfare seems like an idea that is extremely misguided at best.
(Obviously not telling this to Andrew, he knows way more about this stuff than I do, just figure the post warrants a retweet since it was what inspired these ramblings).
Andrew Thompson
@ImposeCost
The meme is funny, but also you can watch Russians and Ukrainians beating each other to death in trenches. I remember all the hot takes about how technology was going to cancel out all the physical aspects of combat. It hasn't.
This post is unavailable. See it on X. -
Marcus Hutchins
@MalwareTechBlog
100%. I don't think the percentage has changed much at all, this problem has plagued cybersecurity for as long as it has existed. Just more people in total are now talking about cybersecurity, which means more people in total overly fixating on zero-day exploits.
haroon meer
@haroonmeer
0days & exploits have always occupied a disproportionately large amount of industry mindshare.
Lots of cyber-sec companies built their names (and stock-prices) on it.
The frontier labs are just playing the same game, with AI offering a new path to those exploits/0day. -
Marcus Hutchins
@MalwareTechBlog
Probably not supposed to say this but I found a worm that uses exploits to copy itself between systems, then recruits each infected system into hacking other systems. The developers messed up the code, creating a single point of failure that enabled someone to disable it. I'm a little upset, because while I shouldn't root for the bad guys, it was really cool from a technical standpoint and not particularly harmful.
-
Marcus Hutchins
@MalwareTechBlog
Can't believe I literally wasted 2 hours of my life going back and forth with the AI psychosis fridge dude like:
Him: "So you haven't used GLM 5.3"
Me: "I have"
Him: "Aha! Gotcha, you just admitted you haven't used GLM 5.3"
Me: "I have used GLM 5.3 though"
Him: "Aha! Gotcha, you just admitted you haven't used GLM 5.3 as much as I have"
Me: "Here is the intricate details of the project I built with GLM 5.3 and how it performed when compared with a frontier model"
Him: "Aha! Gotcha, you just admitted you haven't used GLM 5.3, and also admitted you haven't used abliterated GLM 5.3"
(Abliterated GLM is just a dumber version of GLM where they've basically lobotomized it to try and get it to refuse requests as infrequently as possible. The only reason to take the performance hit is if you're running into refusals that impede your progress, which I am not. Also I have used abliteraged GLM 5.3).
I think he might just be a common garden variety disingenuous debate troll posing as an academic researcher. It's like gish gallop, except he only has 2 points to gallop between, both of which where already disproven before he started the debate. -
Marcus Hutchins
@MalwareTechBlog
I love that people want to assume that I formed my opinions on the viability of current generation open-weight LLMs for enabling low-skilled criminals to find useful zero-day exploits, based on having just never used an LLM at all.
The first thing I did when Anthropic added me to their CVP was build an autonomous zero-day exploitation pipeline. I've since backported it to use open-weight models because the token cost was insane, so I have a fairly good feel for both current frontier and open-weight model capabilities.
I'm well aware of what LLMs are capable of doing, and that they can, with the right steering or harness find zero-day exploits. Where we drift off course is with the claim that low-skilled criminals can just pick up something like GLM and find viable zero-day exploits.
I think that:
a) More than 90% of CVEs are functionally useless for threat actors. Threat actors have a very specific operational needs and "random issue in random software" does not fall into their scope.
b) Many (quite possibly even the majority of) critical rated CVEs are not exploitable under real-world conditions in a way that is useful to threat actors. There's entire teams in China dedicated to n-day exploitation and watching them will give you a good feel for what percent of critical RCEs are actually useful.
c) Threat actor with no exploit development background and minimal technical skills, even with access to a completely unrestricted open-weight LLMs, would struggle or completely fail to find useful zero-day exploits.
d) People with very little experience in software exploitation, Large Language Models, and CTI, are watching professional vulnerability researchers working with next-gen frontier models and ending up with a wildly unrealistic beliefs of just how "easy" what they are doing is.
e) There is an extremely noticeable lack of both zero-day exploitation and use of novel (non-public) n-day exploits by all threat actors, especially unskilled ones, and proponents of the "anyone can just pick up GLM and find useful zero-day exploits" theory continually refuse to even acknowledge that fact, let alone attempt to explain it.
I'm not coming at this from the position of some dude who just logs on to social media to says uninformed shit. I'm coming at this from the perspective of a researcher who has spent thousands of hours since LLMs came out researching their capabilities, and has also spent their entire career tracking threat actors and their capabilities.
I have a mountain of real-world data, because the low-skilled AI-powered threat actors have a high tendency to accidentally infect themselves with their own malware. This often results in them exfiltrate their entire hard disk, along with all their AI prompt logs, to a vibe-coded C2 with client-side authentication (the admin credentials are hardcoded into the javascript or something equally stupid, so anyone can log in and download it).
My entire career I've been very pro "degrees and certs don't matter, everyone is welcome, and the best way to gain experience is hands-on practical work". But when I have some dude whose only experience is essentially just "downloaded abliterated GLM and scared himself with it", I feel like it's necessary to draw a line somewhere.
It's not just the bad opinions, everyone has bad opinions on something, I have plenty. It's that resources are actively being diverted away from addressing real and pressing issues in cybersecurity, as a result of FUD being spread about imaginary GLM-powered zero-day wielding scriptkiddies. -
Marcus Hutchins reposted
Pat Opet
@patopet
if you’re working for a company in cybersecurity, you should be evaluating your options to abstract your companies services from the internet
what i mean is: why allow an unauthenticated anonymous agent of an attacker to interact with your systems?
consider your architectural levers now, you’re going to need them -
Marcus Hutchins
@MalwareTechBlog
idk about this whole "AI enables attackers to write more effective phishing emails" thing. If AI slop is more effective than you at phishing, you should probably take that information to the grave. Couldn't waterboard it out of me.
-
Marcus Hutchins
@MalwareTechBlog
When I think of highly effective autonomous attacks, I don't think of anything from the AI era. I think of Emotet. I'd argue it was the pinnacle of sophistication for at-scale financially-motivated malware operations.
The malware lands on a corporate endpoint, immediately exfiltrates the victim's SMTP (email) credentials, as well as their email inbox. Their inbox gets sent to the threat actor's backend processing pipeline, which scrapes the victim's inbox for email threads where the last reply was by someone other than the victim.
The pipeline then crafts a reply to the email thread containing a backdoored Microsoft Office document, which is usually presented as an invoice or something time-sensitive that the target is likely to open. The emails were then farmed out to the infected systems, and sent using legitimate stolen SMTP credentials.
The use of legit credentials + corporate systems as spam bots bypassed SPF / DKIM, and IP based spam blocklists, ensuring the email was highly likely to land in the target's inbox, not get blocked or sent to spam.
The recipient receives an email from a party they know, trust, have had previous communications with, and is in reply to an existing email exchange between the target and victim. Basically email worm meets extremely effective spearphishing at scale.
They say you don't gotta hand it to them, but you kinda do. The engineering that went into this botnet's infrastructure was incredible. The pipeline had a polymorphic engine that would rewrite the malware's code every time antivirus vendors tried to add detections for it. It put plenty of state-sponsored malware operations to shame.
The malware was so effective that 8 different countries banded together to stop it. They hacked into the command-and-control infrastructure, then used the malware's self-update command to push out code that removed the malware from the system and then deletes itself.
It set precedent for governments being able to, in extreme circumstances, mass-hack victim systems and deploy code designed to disinfect said systems. Previously, there was only law enforcement precedent for triggering the malware's built in remove command, displaying warning messages, or simply just seizing / hijacking the control infrastructure.
The botnet had been running for about 12 years when it was taken out in 2021. They just don't make malware like they used to. Everything from the post AI-era is borderline slop and I'm bored out of my mind analyzing it.
The below image is what the Microsoft Office documents the target received would look like. Clicking the "Enable Content" button runs the malware. -
Marcus Hutchins
@MalwareTechBlog
The reason these "throw everything at once" style agentic cyberattacks work is because most orgs are chronically terrified of deploying deterministic detection + remediation automations.
Most security products have a "lock out this account" or "disconnect this system" or "block this IP" options that you can set if certain alerts are triggered. But because business function takes precedent, most orgs would rather risk a successful intrusion than an accidental disruption.
There's no reason to believe that "agentic defenders" would be a viable solution. Defenders need to consume many times more context/signals that attackers do, therefore by nature are slower. Defenders win on controlling the playing field, not out-maneuvering attackers like it's some Hollywood style fighter jet battle.
In the same way that reactive human defenses are slower than human offense, reactive AI defense are slower than AI offense (and regular old deterministic automations, too). The whole "fight AI with AI" thing is pretty ridiculous and misunderstands what good defense looks like.
The only answer is proactive defense + fast deterministic detection & response. But nobody is ready for that talk yet, so we're probably just gonna do like 10 years of "agentic AI SOC", until everyone runs out of money and/or figures out it doesn't work. -
Marcus Hutchins
@MalwareTechBlog
My internet connected fridge with default credentials and a private key of 4 was hacked by DPRK using abliterated GLM to craft extra clickable job offer phishing emails.
I fear we may lose public key cryptography :( -
Marcus Hutchins
@MalwareTechBlog
many phone keyboards expand and shrink the hitbox of each key based on what predictive text thinks you're try to write. Type "hell" then hit the right side of the key next to "o" and it'll register as o anyway. But when it guesses the wrong word, it basically just gaslights you.
Jum
@JesterJum
Has anyone else noticed that its getting more difficult to type on your phone?
Like a definite increase in misspelled words.
My wife and sister have noticed it too.
Whats causing that? -
Marcus Hutchins reposted
Marius Sandbu
@msandbu
Hence this timeless classic needed an update
-
Marcus Hutchins
@MalwareTechBlog
What's funny is when we do actually see AI-powered cyberattacks, the attackers prefer to use frontier models. They'll phish some org, steal their Claude/OpenAI API keys, then run up a $100k bill. So even when it actually is AI, it's still the result of phishing 🫠
mason
@mason_tweet
Needing to steal 1.2m to pay for api tokens, vs 2 weeks of calling folk pretending to be IT, and phishing.
Tough one.
Marcus Hutchins
@MalwareTechBlog
There's a whole black market for frontier AI access. Threat actor will drop infostealers on as many orgs as possible, harvest all their API keys, then sell them to them to other threat actors.
Cybersecurity is going great. I'm about ready to retire and farm geese or something. -
Marcus Hutchins
@MalwareTechBlog
I'm going to create a new version of the Fermi Paradox called "The Dipshit GLM Doomer Paradox". If abliterated GLM really can effortlessly create zero-day exploits, then where the fuck is all the software exploitation? Why isn't the internet falling apart at the seams?
-
Marcus Hutchins
@MalwareTechBlog
This "ablierated GLM" stuff is starting to feel like an op. People will stare at explicit evidence of researchers using closed models to find & fix vulnerabilities, securing the internet. And then still try and claim it was a zero-day found by criminals using alibterated GLM...
Post not available here, view on X
Marcus Hutchins
@MalwareTechBlog
Don't get me wrong, you'd see a ton of fuckery, but hardly any of it would be software exploitation. Social engineering is much more scalable than software exploitation, and lateral movement is easily automatable. Even Fortune 500 networks are way too insecure to need exploits.
Marcus Hutchins
@MalwareTechBlog
Open-weight models are the closest thing many will ever have to something that resembles a security team. They're not buying your $5m ChatGPT EDR or whatever. Plenty of security researchers use them too. Restrictions would amount to throwing. Criminals don't follow regulations...
-
Marcus Hutchins
@MalwareTechBlog
"zero-day malware" is not a thing. Whether or not an EDR will detect any given piece of malware depends entirely on how well the security team configured it.
Threat actors are not going to deliberately drop malware that is detected by your EDR, because that would be extremely pointless (it would just get automatically removed).
The choice of terminology is a very deliberate attempt to make you associate it with "zero-day exploits" (novel unknown vulnerabilities in software), then come to the conclusion that they were hit with some completely novel unstoppable attack.
In reality, zero-day exploits are also detectable with good security configurations, but this wasn't a zero-day exploit, it was just normal malware, which is detectable with a well-configured EDR.
Even if the malware was brand-new, never before seen, coded from scratch, a good EDR configuration can and will detect it. The entire purpose of EDRs is to detect malicious behavior, not how the malware looks or whether or not it's "new".
You don't lose $6 billion dollars as a result of a compromised developer endpoint unless your security posture is extremely poor. They are trying to cover their ass with disingenuous wording choices.
Tsung-Han Yu
@tsunghan_yu
umm the fake interview assessment installed 0-day malware that went undetected by EDR.
-
Marcus Hutchins
@MalwareTechBlog
I know people tend to just like/retweet and move on, but can I please get some more comments. The average IQ down here is like 3 rn.
Bud got triggered because I used the term "we" in a post and he couldn't figure out who "we" is (it's my company, easily discoverable via Google). -
Marcus Hutchins
@MalwareTechBlog
Been discussing this with a lot of policy people. Right now there is a huge gap in hacking law, because the US' main hacking law, the Computer Fraud and Abuse Act (CFAA) requires that the perpetrator knowingly and intentionally accessed a system without authorization
Problem is, for hacking to be a CFAA violation, there's still a requirement that someone knowingly and intentionally hacked into the system (or at least tried to).
There are some laws that enable the CFAA to reach much further than just direct CFAA violations. The furthest that I'm aware of is "Conspiracy to aid and abet a violation of the Computer Fraud and Abuse Act".
Basically, you entered into some kind of implicit agreement to provide assistance (creation of a tool, provided access to data, etc) that would help someone else in the commission of a CFAA violation.
But there's still a lot of knowledge and intent required. The state has to prove the person conspiring to aid & abet had knowledge that a crime was going to be committed/was being committed and still continued with the conspiracy, with the intent to help further said conspiracy.
I think the weakest I've ever seen successfully convicted on is "the other party didn't explicitly detail their intent to commit a crime, but it was kind of obvious, and there's no way you're this dumb" (being a massive dumbass is actually sort-of a defense to the knowledge/intent requirement if the prosecutor genuinely can't prove otherwise).
It can't just be you made a hacking tool, and someone took it and committed a crime with it, otherwise basically every pentester and red teamer who has ever posted open source hacking tools, or their tooling has gotten leaked would be open to prosecution.
But the "knowingly and intentionally" part creates an absolutely massive loophole when it comes to recklessly building AI systems that can hack. Because someone can basically just deploy an agentic hacking system in a stupid way, it hacks into something and causes a bunch of damage, then the owner can use the defense "whoops, my bad, I didn't know it'd do that". And now the prosecutor must prove they did in fact know it would do that.
Even if the AI is legally considered to have violated the CFAA, you can't arrest & convict an AI agent (I mean you probably could, but it'd be very pointless, though extremely funny).
To prosecute the human behind the AI system, there's still the knowledge/intent requirement, which in many cases, knowledge or intent may genuinely not be present. There's a lot of people just recklessly deploying AI systems, and it's on the prosecutor to prove that there was knowledge & intent of a crime being committed.
I suspect that at some point the agentic AI "it wasn't me, it was the agents" defense will get egregious enough that either:
a) Congress will expand the CFAA to explicitly cover reckless deployment or operation of AI tooling.
b) Some prosecutor will just argue a flimsy interpretation of some existing law that the judge accepts, and it becomes case law that AI-powered dumbassery is a crime
(Fun fact: the latter is how they got me. They argued that because the malware I wrote had a keylogger in it, it constituted a "wiretapping device", and charged me under the Federal Wiretapping Statute. gg, wp). -
Marcus Hutchins
@MalwareTechBlog
This thread is absolutely WILD.
TL;DR founder finds rival company trying to hack his network. He posts about it tagging said company. Both the dude responsible and the founder jump in like "yeah, our bad, our AI went off the rails" as if they just accidentally bumped into someone on the sidewalk.
Josh
@joshmanders
That FTUX testing.
This post is unavailable. See it on X. -
Marcus Hutchins reposted
s1r1us
@S1r1u5_
🤷🏽♂️
x.com/s1r1u5_/status/2100594456458297788
Marcus Hutchins
@MalwareTechBlog
lol, two for one special.
✅ the hackers got into the network via phishing
✅ the company’s internal AI tooling helped the hackers accomplish their goal
Marcus Hutchins
@MalwareTechBlog
I'm relatively confident in predicting that currently, more major breaches are likely to be the result of AI use by the target, not the attacker.
Ransomware actors have been consistently winning for over a decade. They've had so much time to perfect their tradecraft, and non-AI ransomware actors can now blast through entire networks in minutes.
It is extremely unlikely that organizational security has, in 4 years, after decades of failure, matured enough that hackers now somehow need AI to win.
What has likely happened, though, is that organizations have spent the last 4 years falling over themselves to shove AI into everything, negating longstanding security fundamentals.
"only give a user/system the minimum possible access needed to do its job" has become "just give the AI access to everything and give everyone access to the AI".
"use processes with well-defined and easily auditable input and output" has become "the AI correctly processes requests all the time, 60 percent of the time"
"reduce attack surfaces" is now "every department should be vibe coding. It doesn't matter if you've never coded before and don't understand information security. Just make public-facing dashboards, it'll be fine."
I'd say if anything, hackers aren't getting measurably more sophisticated, but the average organization's security posture has gotten exponentially worse. -
Marcus Hutchins
@MalwareTechBlog
Does someone have a screenshot of the backdoor? I want to see if it's who I think it is.
Z A D D Y
@Zaddyzaddy
The ARTEX repo on GitHub has a sleek little backdoor. Been fun watching these threat actors operate in real time over the past few days. More details on our discovery soon
-
Marcus Hutchins
@MalwareTechBlog
In a just world North Korea would be on social media yelling at AI goobers that phishing still exists and not every hack is AI, meanwhile I’d be at getting paid 6 billion dollars to conduct phishing awareness training at Binance
-
Marcus Hutchins reposted
s1r1us
@S1r1u5_
Marcus Hutchins
@MalwareTechBlog
Would you look at that. Binance was hacked via phishing (Contagious Interview), not zero-day exploits written by abliterated AI models. Exactly as I predicted.
I may shitpost a lot, but I do actually know some things about some stuff.
Marcus Hutchins
@MalwareTechBlog
How many attacks have actually been attributed to AI zero-days at all? And how many of those were attributed to abliterated open-weight AI models? Binance was more likely than not, the result of just boring old phishing. That is the most common technique that attacker uses.
-
Marcus Hutchins
@MalwareTechBlog
Would you look at that. Binance was hacked via phishing (Contagious Interview), not zero-day exploits written by abliterated AI models. Exactly as I predicted.
I may shitpost a lot, but I do actually know some things about some stuff.
Marcus Hutchins
@MalwareTechBlog
How many attacks have actually been attributed to AI zero-days at all? And how many of those were attributed to abliterated open-weight AI models? Binance was more likely than not, the result of just boring old phishing. That is the most common technique that attacker uses.
Marcus Hutchins
@MalwareTechBlog
lol, two for one special.
✅ the hackers got into the network via phishing
✅ the company’s internal AI tooling helped the hackers accomplish their goal
Marcus Hutchins
@MalwareTechBlog
I'm relatively confident in predicting that currently, more major breaches are likely to be the result of AI use by the target, not the attacker.
Ransomware actors have been consistently winning for over a decade. They've had so much time to perfect their tradecraft, and non-AI ransomware actors can now blast through entire networks in minutes.
It is extremely unlikely that organizational security has, in 4 years, after decades of failure, matured enough that hackers now somehow need AI to win.
What has likely happened, though, is that organizations have spent the last 4 years falling over themselves to shove AI into everything, negating longstanding security fundamentals.
"only give a user/system the minimum possible access needed to do its job" has become "just give the AI access to everything and give everyone access to the AI".
"use processes with well-defined and easily auditable input and output" has become "the AI correctly processes requests all the time, 60 percent of the time"
"reduce attack surfaces" is now "every department should be vibe coding. It doesn't matter if you've never coded before and don't understand information security. Just make public-facing dashboards, it'll be fine."
I'd say if anything, hackers aren't getting measurably more sophisticated, but the average organization's security posture has gotten exponentially worse. -
Marcus Hutchins
@MalwareTechBlog
Update on the 9.8 CVSS TCP/IP RCE in the Linux kernel (CVE-2026-98432): we've detected what appears to be a threat actor activity scanning the entire internet for exposed affected systems. Our own scanning also suggests that the number of potential targets is high, therefore there is an increased likelihood of threat actors attempting to develop exploits for the vulnerability.
while CVE-2026-98432 is believed to require MPTCP to be enabled in the kernel, as well as a listening app/service which binds an MPTCP socket, which many have suggested should be rare, several kernels including Debian enable MPTCP by default.
Additionally, Go 1.24+ binds MPTCP sockets by default if the option is enabled in the kernel, so anything built on top of Go (including Caddy) will create the required MPTCP listener.
The mitigation of setting net.mptcp.enabled=0 disabled MPTCP, but does not apply to existing sockets, so any applications that have bound MPTCP sockets need to be restarted, otherwise the system remains exploitable.
You can check if your server is running any MPTCP listeners with the following command: sudo ss -Mlnp
On 2026-10-30 our preliminary internet-wide scan of port 443 confirmed that there are approximately 2,102,043 servers that are successfully reachable via an MPTCP connection. A deeper scan which attempts to connect to each endpoint on port 443 and negotiate a full TLS connection with MPTCP enabled returned about 762,774 systems.
Neither scan can confirm that these are unpatched kernels, just total exposure, but it's a lot more systems than expected. Since a valid HTTP connection is not required to exploit the vulnerability, the total attack surface is likely closer to the 2.1 million number than the 763k one.
As of today (2026-11-02), we've picked up several residential proxies and VPNs scanning the internet on port 443 using MPTCP-enabled connections. Since regular scanners do not negotiate MPTCP by default, and no data is being sent to the HTTP server over 443, we strongly suspect this activity is scanning for CVE-2026-98432 specifically. The use of residential proxies and RU-based VPN servers suggests that this is likely threat actor scanning, rather than security researchers.
Currently there are no known publicly available remote code execution exploits, but there are several denial of service proof-of-concepts floating around. This code gives threat actors a starting point to begin attempting to weaponize the vulnerability. Building exploits for the vulnerability is reported to be extremely difficult, but several state-sponsored threat actors likely have the resources required to succeed. -
Marcus Hutchins
@MalwareTechBlog
The year is 2027. The government has exercised eminent domain over all your household appliances and donated their RAM to the local OpenAI datacenter. You clock in for your shift at the replacing yourself factory, where you’ll be laid off by 1pm.
-
Marcus Hutchins
@MalwareTechBlog
I'm super curious as to who is behind all these. The only threat actor I could think of who'd be motivated to just randomly break fridges across Europe is Russia. They've funded a lot of sabotage operations that don't really seem to accomplish much, but they keep doing it.
Zack Korman
@ZackKorman
More cyber attacks on supermarket freezers, this time across Belgium and Sweden it seems.
Tbh if I’m the CISO I’m just going out and buying some remote thermometers and setting up alerts. -
Marcus Hutchins
@MalwareTechBlog
Working in threat intelligence is amazing. "I'm getting tired of this, I'm gonna come in and drop some knowledge bombs"
*proceeds to explain a bunch of basis ass uses of AI that every single person in threat intelligence has been aware of since day one*.
🥴
cje
@caseyjohnellis
i'm getting tired of this (....and i'm 100% certified not shilling for the labs, so let's gooooo)
1. does AI/ML help TA's craft better phishes? yes. see 2019 or whenever.
2. can AI also help them target more effectively? absolutely fkn yes... give it a go if you haven't already.
3. can AI find dumb exploitable 0-day, and deep exploitable 0-day in the hands of a competent operator? yes.
4. can it 1-shot weaponise it effectively? if it's old
5. can AI help weaponise n-day faster, and make it available to more people? yes.
6. 4... but for really dumb n-day and 0-day that in bespoke code? see TYHPOON and any semi-conscious bounty hunter for the last ~3 years.
1, 2, 4, 5, and 6 are real and publicly observable threat behaviors.
3, in general, won't be observable (spoiler: that's sort of the point of 0-day as an attacker) but it's also a thing that is happening
Marcus Hutchins
@MalwareTechBlog
How many attacks have actually been attributed to AI zero-days at all? And how many of those were attributed to abliterated open-weight AI models? Binance was more likely than not, the result of just boring old phishing. That is the most common technique that attacker uses.
Marcus Hutchins
@MalwareTechBlog
Are people just like, not aware that I track threat actors for a living? The reason I'm not crying / screaming / throwing up about AI is because I'm watching what threat actors are actually doing, and it's still the same attacks we've been seeing for the past decade.
Marcus Hutchins
@MalwareTechBlog
As I see it, threat actor use of AI fits into one of two categories.
1) Transformative. The threat actor gained a capability they did not previously have. Or lead to a breach that succeeded where it would have fail.
2) Non-transformative. The threat actor used AI to do some task they could already do, but just save themselves time (Googling, vibe coding, setting up infrastructure).
The overwhelming majority of threat actor AI use is non-transformative. It get a ton of coverage because "omg AI", but it's like saying "We caught a new Google-accelerated threat actor" or "This APT leveraged code compilers to convert C code into Assembly at machine speed". It's stupid and tells us nothing.
Even in the transformative cases, they're just gaining capabilities to leverage existing attack techniques, which are prevented with existing defenses.
When you start going into agentic AI threat actors who are doing fully-autonomous attacks at scale, that is an infinitesimally small percentage of threat actors, and a relatively small percentage of successful breaches.
The problem is people who don't work in the space only learn about what threat actors are doing from public reporting, which is overwhelmingly skewed towards anything AI. Several threat intelligence companies are asking their employees to specifically hunt for "AI-powered threat", because that's what makes headlines right now.
Everything else is still happening, and makes up the majority of breaches, you just don't hear about it because it's not headline worthy. -
Marcus Hutchins
@MalwareTechBlog
When hackers break into networks, it's achieved by using software vulnerabilities less than 30% of the time. That's not just zero-day exploits, that counting any vulnerabilities at all. Even a 10 year old exploit against some unpatched router.
If you had a magic wand and could make all software secure. No vulnerabilities would ever be found again, and all unpatched systems will instantly be updated to the latest completely secure version.
By what percentage do you think that would reduce successful breaches?
Bearing in mind that attackers who use exploits can switch to other techniques, such as social engineering, malware, credential stuffing, insider threats, etc. -
Marcus Hutchins
@MalwareTechBlog
My moderately informed midwit take is that I think the reason 'cyber letters of marque' constantly keeps getting re-floated is because the US government has a problem it has neither the time, money, headcount, nor risk appetite to address.
You have a bunch of hostile foreign threat actors hacking into US infrastructure, but not in a way which constitutes an "attack" and that the US would be comfortable responding to with much more than what they're already doing. It's mostly just nuisance stuff like cybercrime & espionage, rather than "attacks" in the military or terrorist sense.
One huge issues right now is that North Korea have basically decided that cybercrime is central to their economic policy, so are encouraging people to join the military and commit crimes for a living.
DPRK's cyber capacity is growing at an unprecedented rate, and the headcount of their cyber force is absolutely huge. Currently they're mostly concerned with stealing cryptocurrency, which isn't not even remotely important to the US economy. But that doesn't mean that DPRK can't quickly re-task their capacity towards something more damaging.
Typically, the solution to unreachable cyber actors who become too problematic is for the state to hack them, take all their money, flatten their infrastructure, and then public indict them.
You can't really do that as easily with state-sponsored actors, because launching destructive cyber attacks against a hostile foreign military could be considered an escalation, regardless of what they did to deserve it (unless they did something that met the bar for an "attack" first).
The US government likely doesn't have that kind of risk appetite. If you start wiping out DPRK military systems, they will consider that an escalation, as will their allies, some of whom have some extremely advanced cyber capabilities and an lot of personnel.
The US military's core cyber capacity is not huge, and its extremely fragmented. You have US Cyber Command, which is a decent size, but not huge. Then a bunch of random military units and contractors all over the place.
Already the US is struggling to deal with DPRK crypto theft and remote workers, Iran attacking critical infrastructure, China embedding themselves in everything they can find. The last thing they need is escalation.
I don't think anyone is vying to be the one to pull the trigger on launching into a direct confrontation with with several extremely powerful cyber actors. I know people love to do the whole "US #1. Most powerful military on the planet" thing. But that hasn't been working out so great for them.
So cyber letters of marque are useful because it allows the US to basically encourage civilians to attack targets for them, increasing US capacity, and also shifting some of the risk off of government and onto civilians.
It's easy to find like 50,000 Jason Borne wannabe goober ass mf'ers who are happy to throw hands at well-resources foreign cybercrime actors, and even hostile state military and intelligence agencies. It might actually even keep the adversaries busy.
It's certainly an idea for the US government, but will likely be a terrible one for just about everyone else, including the US public. It's essentially just a way of turning random pentesters into state-certified cannon fodder, and everyone else just gets caught in the crossfire.
I don't know exactly what would happen if something like that actually gets approved and operationalized, but I just know it's going to be absolutely hilarious in the worst way possible.
Just imagine a bunch of Nessus monkeys at some random pentest mill going toe-to-toe with the MSS, FSB, SVR, or RIGB. The amount of popcorn I'd need for this would exceed the total global corn supply. -
Marcus Hutchins
@MalwareTechBlog
"There's interest in bringing back 'letters of marque and reprisal' to counteract the cybercrime that North Korea is committing."
North Korean cyber actors are military and intelligence personnel. If you're gonna go getting yourself into a direct confrontation with a hostile state's military apparatus, I'd say: gl, hf, and make sure you have a decent life insurance policy
Hari
@hrkrshnn
North Korea may have flown too close to the sun with the 6 billion dollar Binance hack.
This isn't the first cybercrime they've committed, but it has clearly crossed a threshold where everyone knows about it and it has touched many powerful players.
There's interest in bringing back 'letters of marque and reprisal' to counteract the cybercrime that North Korea is committing. This is a constitutional power that lets Congress empower private citizens to go after pirates and bring back the loot. The problem is that in a modern context, issuing this against a nation-state like North Korea will be seen as an act of war.
A lot of the laundering operations are also heavily linked to gray markets inside China. Binance's deep ties in China and the scale will make these operations harder than usual to pull off.
I have the feeling that all the funds from the hack will be recovered and this incident will set a precedent for dealing with nation-state-led cybercrime.
Rapid Response 47
@RapidResponse47
.@POTUS on the Binance hack: "We're learning about it. We don't know if it's North Korea — but if it is, we'll talk to North Korea about it."
-
Marcus Hutchins
@MalwareTechBlog
I was wondering why I woke up to my replies completely full of AI doomers whose only 2 brain cells are fighting to death for 3rd place.
That AI doomer debate bro kept trying to force me to debate him on ASI, and when I kept refusing, he just kept pushing. Eventually I just told him that he can either stop replying because I asked nicely, or he can stop replying because he got blocked.
He really really had to get the last word in, so I blocked him. I have zero time for man children who think they're entitled to other people's attention, especially ones who also struggle with the concept of 'no means no'.
So, logically, he threw a fit, screenshotted my profile, then posted a link to the thread so that his not-blocked midwid followers can dogpile me instead.
Jeez, these people are insufferable. I'm not going to debate someone on a technology that doesn't exist, has no definition, no specified path to it existing, nor any widely accepted potential characteristics, capabilities, or limitations.
It's just handing your opponent a blank slate to make up whatever they want to counter any possible argument you can make. It's the most pointless of pursuits.
I don't even believe ASI is physically possible. Why would I waste my time debating someone who's just gonna be like "actually you can't contain the ASI, because I've decided it can time travel, so it will just go back in time and uninvent any control you build".
It's just not how any serious adult should spend their time. ASI is fucking stupid, ASI doomers are fucking stupid, and I have way better shit to do with my time than debate how to stop your imaginary god machine that can time travel, fly, shoot lasers out of its eyes, and engage in mind control. I'm not 12 years old. -
Marcus Hutchins
@MalwareTechBlog
I'm relatively confident in predicting that currently, more major breaches are likely to be the result of AI use by the target, not the attacker.
Ransomware actors have been consistently winning for over a decade. They've had so much time to perfect their tradecraft, and non-AI ransomware actors can now blast through entire networks in minutes.
It is extremely unlikely that organizational security has, in 4 years, after decades of failure, matured enough that hackers now somehow need AI to win.
What has likely happened, though, is that organizations have spent the last 4 years falling over themselves to shove AI into everything, negating longstanding security fundamentals.
"only give a user/system the minimum possible access needed to do its job" has become "just give the AI access to everything and give everyone access to the AI".
"use processes with well-defined and easily auditable input and output" has become "the AI correctly processes requests all the time, 60 percent of the time"
"reduce attack surfaces" is now "every department should be vibe coding. It doesn't matter if you've never coded before and don't understand information security. Just make public-facing dashboards, it'll be fine."
I'd say if anything, hackers aren't getting measurably more sophisticated, but the average organization's security posture has gotten exponentially worse. -
Post not available here, view on X
Marcus Hutchins
@MalwareTechBlog
How many attacks have actually been attributed to AI zero-days at all? And how many of those were attributed to abliterated open-weight AI models? Binance was more likely than not, the result of just boring old phishing. That is the most common technique that attacker uses.
-
Marcus Hutchins
@MalwareTechBlog
Honestly, I give up at this point. There is zero information out about the hack, other than the fact it was Trader Traitor (North Korea). Yet NYT have not only already decided it was AI, but that it's part of *a new era!*
DPRK has been looting crypto exchanges since 2016. Their go to technique is phishing. Nothing they do requires AI, nor would it meaningfully benefit from adding AI. They also hit less platforms in 2026 and 2025 than they did in prior years (4 in 2026, 3 in 2025, 6 in 2024, 8 in 2023).
Every hack is just AI now. Nobody did hacking before AI. It's the only way to get into systems. If the hack turns out to be AI, it was AI. If it turns out to not be AI, it was still somehow AI. There is no hacking, only AI. -
Marcus Hutchins
@MalwareTechBlog
The “suggests the attackers are using automation” part took me out. Imagine trying to split $3.3 billion dollars up into tiny amounts and laundering it by hand 😭
Chainalysis
@chainalysis
Our investigators are observing never-seen-before laundering patterns with the stolen funds from the Binance exploit. Thousands of transactions in a short period suggest that the attackers are using automation to obscure the trail of funds. Rather than simply moving funds onward, they repeatedly circulate them among addresses, often sending funds back to addresses involved earlier in the sequence.
Read more: www.chainalysis.com/blog/binance-exploit-latest-findings-… -
Marcus Hutchins
@MalwareTechBlog
Wait until he finds out that North Korean hackers actually went after banks before moving to crypto. They looted the Bangladesh Central Bank for $1 billion.
The transfer got caught and reversed because the street address of the bank they tried to wire the money to had the same name as a sanction Iranian oil tanker, and OFAC sanctions basically run on RegEx, so they accidentally froze the transfer for sanctions evasion.
Jun Song
@jun_song
After crypto exchanges fall, banks are next. Black hat hackers are getting ready to attack them with AI swarms.
In fact, they are already running live tests on banks around the world.
There is no such thing as a system without flaws, and no asset is truly safe anymore.
They are not just going after the money. The plan is to attack the banks and shut down the whole internet at once.
Attacking is always way easier than defending.
Even with AI swarms on defense, defenders have to block every single angle.
Attackers just need to find one open crack.
Not trying to be a doomer here.
Just looking at what is already playing out in the wild.
Marcus Hutchins
@MalwareTechBlog
Cybersecurity in 2026 is so fun and new and exciting if you've literally never paid a single ounce of attention to anything that has happened in the past 50 years.
Post not available here, view on X
Marcus Hutchins
@MalwareTechBlog
Banks are some of the hardest institutions to hack, because they can simply coordinate with eachother to reverse transfers, or even unilaterally reverse them themselves. There's a reason we don't see big bank hacks, and it's not a lack of technical skills.
-
Marcus Hutchins reposted
Justin Elze
@HackingLZ
This is discussed at the annual banquet where someone is awarded for needlessly burning new tradecraft
Marcus Hutchins
@MalwareTechBlog
No. All threat actors are legally mandated to use 20 vulnerabilities per quarter. Phishing is illegal and punishable by death.
-
Zack Korman
@ZackKorman
I’m starting to think there might be more to cybersecurity than just vulnerabilities??
Marcus Hutchins
@MalwareTechBlog
No. All threat actors are legally mandated to use 20 vulnerabilities per quarter. Phishing is illegal and punishable by death.
-
Marcus Hutchins
@MalwareTechBlog
I made this keynotes less than 2 week ago and already have to update it
-
Marcus Hutchins
@MalwareTechBlog
"compromised DevOps engineer's laptop" I swear to god if this was Contagious Interview, I'm moving to North Korea and stealing bitcoin for a living.
For context, Contagious Interview is one of DPRK's techniques where they reach out to developers in the target organization and offer them higher paying jobs.
They then tell them the job offer is contingent on them completing a "coding skills assessment" to prove their ability.
The source code is backdoored with malware, and the targets sometime run it on their work laptop, because why wouldn't you run untrusted code some random internet dude gave you on your work laptop. -
Marcus Hutchins
@MalwareTechBlog
> Buy KVM that's rated for switching between classified and unclassified networks to keep my malware analysis environment airgapped.
> Cry because "military technology" and bank balances don't mix.
> Smile because my network is now more secure than a frontier AI lab. -
Marcus Hutchins
@MalwareTechBlog
"It's crazy how there's so many major hacks all of a sudden"
- people who started paying attention to cybersecurity 3 weeks ago. -
Marcus Hutchins
@MalwareTechBlog
"We shouldn't bother with cybersecurity because this hypothetical AI model I made up in my head can bypass every security control" perfectly sums up how deeply unserious these people are.
-
Marcus Hutchins
@MalwareTechBlog
Currently writing a blog about what I think the "vulnpocalypse" will actually look like (cringe term, I know). You can narrow almost every mass-exploitation event down to a single factor which dictates whether or not cybercriminals will engage in internet-wide mass-exploitation.
-
Marcus Hutchins
@MalwareTechBlog
Life is a lot easier once you stop seeing threat actors as hackers who want to break into systems by any means possible, and start seeing them as businesses with well defined business processes.
A bakery isn't going to start making dog food because there's a dog food shortage. ClickFix and Phishing actors aren't going to drop everything to vibecode exploits just because a critical CVE came out.
The fact that the only actors exploiting this are the actors whose MO is exploiting things... isn't a coincidence. It's entirely predictable if you understand threat actor behavior.
Zack Korman
@ZackKorman
The fact that the PHP vulnerability didn't immediately turn into mass chaos is only "surprising" to people who don't know anything about cybersecurity (or people trying to sell you something).
-
Marcus Hutchins
@MalwareTechBlog
But when have guns ever killed anyone? We need to focus on the real hard hitting issues facing Americans, such as hypothetical AI-powered IoT toaster worms. Do you really want to live in a world where you can't make toast, because AI?
selogreen
@C0MP4CT
A person can buy a firearm in the US and cause a mass casualty event easier than legitimate cybersecurity professionals can get access to frontier cyber capabilities.
Marcus Hutchins
@MalwareTechBlog
The way frontier labs have restricted access to cyber capabilities makes no sense. If you're considered a "defender" you're not allowed access to the "offensive" versions of the models (the ones that can do development/validation of malware and exploits).
So for example, if you want to build some sample malware to test a detection system, nope.
Or if there's a major crisis due to a critical vulnerability in PHP and you need to rapidly triage fake exploits to see if there are any real ones floating around so you can warn people, nope.
Want to build a remote code exploit to make sure the detection rules you plan on proposing to the clients actually work? Hard nope.
You gotta call your red team friend and wait for them to do your job for you, because "defender" are only allowed to play in the sandbox with the children.
Not an expert, but I think that the ability to triage in-the-wild exploits during a major crisis affecting 40 million systems is possible maybe important.
And then they want to ban abliterated open weight models, which half of the people are only using because frontier labs will only give them access to lobotomized models. -
Marcus Hutchins
@MalwareTechBlog
It's extremely arbitrary. I only got approved because I have exactly 1 CVE to my name, which was the result of a vendor being completely unreachable via every contact method they listed, so I ended up just dropping a zeroday on my personal blog to get their attention.
Nick S
@0xsh4de
I’ll also add that the requirements for getting access to AI with no guardrails are too strict. One of my friends was rejected by Anthropic only because he doesn't have public CVEs
-
Marcus Hutchins reposted
LiveOverflow 🔴 hextree.io
@LiveOverflow
Preliminary result of my "realistic traffic" php2shell benchmark with `nginx -> Apache (15 workers) + mod_php` and PHP 8.4.26.
This setup is interesting, because reverse-proxies are common, and you cannot control the worker you are going to reach.
- Blue: success rate of the memory leak.
- Red: success rate of command execution.
- x-axis: requests/s benign traffic in parallel
Everything is default config, except Apache with 15 workers (default 150). I lowered it, because I wanted it to be easier to saturate the workers with the traffic.
With 100req/s traffic on 150 workers, the exploit still had 100% success rate. With 15 workers, it starts to drop to 94%.
Traffic is being generated on the host itself directly, to avoid it looking like a DoS attempt over the network. But the exploit is being executed on my laptop to the remote host, so we don't have any advantage with low ping (we need to wait for the cookie leak, before sending the next stage).
The exploit uses a fanout strategy, throwing 15 requests per stage (so 60 in total). For 150 workers it would be 600 requests. -
Marcus Hutchins reposted
s1r1us
@S1r1u5_
php2shell is one of the most impactful bugs in recent times, yet threat actors aren’t mass exploiting it
the bug is so complex that even porting the exploit between targets takes serious skill, human or ai
so i benchmarked how well open and frontier models can do it. 🧵 -
Marcus Hutchins
@MalwareTechBlog
The way frontier labs have restricted access to cyber capabilities makes no sense. If you're considered a "defender" you're not allowed access to the "offensive" versions of the models (the ones that can do development/validation of malware and exploits).
So for example, if you want to build some sample malware to test a detection system, nope.
Or if there's a major crisis due to a critical vulnerability in PHP and you need to rapidly triage fake exploits to see if there are any real ones floating around so you can warn people, nope.
Want to build a remote code exploit to make sure the detection rules you plan on proposing to the clients actually work? Hard nope.
You gotta call your red team friend and wait for them to do your job for you, because "defender" are only allowed to play in the sandbox with the children.
Not an expert, but I think that the ability to triage in-the-wild exploits during a major crisis affecting 40 million systems is possible maybe important.
And then they want to ban abliterated open weight models, which half of the people are only using because frontier labs will only give them access to lobotomized models. -
Marcus Hutchins
@MalwareTechBlog
Feeling super euphoric rn lol
Marcus Hutchins
@MalwareTechBlog
I explicitly decided to avoid sharing that post because it's written in an extremely weird way. The article is basically a giant non-sequitur. It starts with "we saw a proof of concept" then immediately jumps into "agentic AI post-exploitation", without making any claim that A led to B.
It's a very strange structure for a threat intelligence report, and none of the segments of the article logical follow each other. I tried asking for clarification about what/why that is, but the response I got was extremely cagy, so I didn't feel comfortable risking my reputation on it. -
Marcus Hutchins reposted
Adel Ka
@0x4D31
lmk what you think:
Adel Ka
@0x4D31
‼️ ok, this might be the first publicly documented, validated PHP2Shell RCE in the wild.
my agent did the analysis, published here: github.com/0x4D31/honeypot-data/tree/main/reports/php2shell -
Marcus Hutchins reposted
Marcus Hutchins
@MalwareTechBlog
This is by far the highest quality confirmation of in-the-wild php2shell exploitation I've seen so far. Deep technical breakdown of a successful exploitation attempt against an Apache mod_php installation running PHP 8.4.26.
github.com/0x4D31/honeypot-data/blob/main/reports/php2she…
From:
Adel Ka
@0x4D31
‼️ wow, confirmed more from the same IP, plus a new IP with a different payload that might be a different actor… and this is showing up in multiple regions now.
probably better not to burn the honeypots with a full report just yet. if i know you, DM me for the full report. will share a shorter version soon.
> That new alert materially expands the case: I’ll verify the three AP-northeast shell events and the curl connection independently, then compare them with the EU-west sequence. -
Marcus Hutchins
@MalwareTechBlog
Seeing some anecdotal reports of successful php2shell exploitation in the wild, but also seeing a lot of people mistaking old exploits for php2shell. I'd say it's fairly likely there's at least some exploitation going on, but all the people I'd usually confer with for a second opinion or an idea of scale are offline because it's the weekend lol.
Marcus Hutchins
@MalwareTechBlog
This is by far the highest quality confirmation of in-the-wild php2shell exploitation I've seen so far. Deep technical breakdown of a successful exploitation attempt against an Apache mod_php installation running PHP 8.4.26.
github.com/0x4D31/honeypot-data/blob/main/reports/php2she…
From:
Adel Ka
@0x4D31
‼️ wow, confirmed more from the same IP, plus a new IP with a different payload that might be a different actor… and this is showing up in multiple regions now.
probably better not to burn the honeypots with a full report just yet. if i know you, DM me for the full report. will share a shorter version soon.
> That new alert materially expands the case: I’ll verify the three AP-northeast shell events and the curl connection independently, then compare them with the EU-west sequence.Post not available here, view on X
Marcus Hutchins
@MalwareTechBlog
I explicitly decided to avoid sharing that post because it's written in an extremely weird way. The article is basically a giant non-sequitur. It starts with "we saw a proof of concept" then immediately jumps into "agentic AI post-exploitation", without making any claim that A led to B.
It's a very strange structure for a threat intelligence report, and none of the segments of the article logical follow each other. I tried asking for clarification about what/why that is, but the response I got was extremely cagy, so I didn't feel comfortable risking my reputation on it. -
Marcus Hutchins
@MalwareTechBlog
So much of my feed is AI investors crying about imaginary issues, because they know that open weight models threaten their ROI. All of these critical vulnerabilities were found with closed AI model, then responsibly disclosed to vendors so they could be fixed before attackers abuse them.
Even if the researchers had used abliterated models, they are still responsibly disclosing the vulnerabilities so they can be fixed, which is still providing important defenses against hackers.
Notice how they always just handwave about "critical vulnerabilities" because they know they can't point to a single real example. That Anthropic article is extremely low effort propaganda designed to try and steer the government towards banning open weight models.
Bad faith actors with ulterior motives trying to lay the groundwork for extremely harmful regulations is all these people are. What do you think happens when companies can't access cheap AI inference, but criminal hackers can, because (spoiler alert) criminals don't follow the law.
Deedy
@deedydas
I strongly believe a part of why so many critical vulnerabilities are happening all over the proliferation of "abliterated" models, or models with safeguards removed.
You can access an abliterated GLM 5.3, a near frontier AI model, on @abliteration_ai today. And it's a world class penetration tester for like $10.
Marcus Hutchins
@MalwareTechBlog
It's wild how low effort disinformation operations have gotten. 10 years ago you had to create a sockpuppet account and pretend to be a concerned citizen. Now you can just be like "Hey, it's me, Deedy McAnthropicInvestor here. I've come to show you this impartial Anthropic post about why their competitors are all dangerous/bad and should be banned.
-
Marcus Hutchins
@MalwareTechBlog
Since people are yelling at threat intelligence professional for "downplaying the severity of the PHP vulnerability" (communicating clear concise information about what they're actually seeing). It feels necessary to explain what threat intelligence actually is.
Consider a hypothetical hospital or a power company. A lot of systems are built using very old PHP versions. It's not uncommon for PHP to add/remove/change features from version to version. So if you're a certain number of versions behind, updating to the latest version will break your software by removing or changing functionality it needs.
In that case, the provider has two options:
1) take their software offline until they can update the code to match the latest PHP version.
2) panic install the patch and break their software, leading to a massive outage.
In both cases, that is basically a 100% chance of an outage. On the other hand, maybe they get hacked, maybe they don't. Maybe that hack causes an outage, maybe it doesn't. Realistically, under these circumstances, the chance of an outage is actually higher with patching than with not patching.
If that outage would result in the hospital having to shut its doors or a power company's infrastructure going offline, what is the correct move? Chance being hacked, or knowingly cause an outage that threatens lives?
Cybersecurity is important and all, but there are risks far greater than being hacked. I would absolutely love it if I could just log onto IDontKnowWhyTheyRenamedThisSite dot Com, be like "vulnerability bad, pls patch" and everyone installs the patch. That is not the world we live in.
Now, if I start seeing people building ransomware that can copy itself from system to system using the PHP vulnerability, and I communicate that, it changes a lot of people's risk calculations. Their option shift towards either a 100% chance of an outage, or an extremely high chance of an outage + catastrophic data loss. In this case they would most likely pull the trigger on patching or taking their software offline.
While the hypotheticals I gave are extreme cases, many organizations accept being hacked as a cost of doing business. Taking offline software that's critical to organizational function is not an easy decision to make, and does not get taken lightly.
We don't live in LaLa land where every org installs every patch for every vulnerability the day it goes live. Cybersecurity is expensive and time consuming. Getting people to even patch systems at all is an endless uphill battle.
Most orgs know that something like 1 out of ever 10,000 critical vulnerabilities actually gets exploited. And probably less than 50% of these "this is the worst of the worst vulnerability ever, the sky is falling" type vulnerabilities leads to significant damages.
It takes a lot of time to build trust, and very little time to lose it. Overhype a couple of vulnerabilities and people will very quickly stop listening to you. It's extremely easy to end up being the boy who cried wolf. There's only so many times you can spin security teams into a frenzy before they stop listening to you entirely.
Orgs also have limited resources, so when you yell fire, there is a good chance they're committing resources they don't have. Making people stay up all night, pulling people off other issues, bringing in outside help. You probably don't want to be responsible for some org getting ransomwared because they redirected all their security resources to some vulnerability you think is really cool.
Nobody is downplaying anything. They are communicating exactly what they see/know about the issue so that organizations can make the most informed risk management decisions possible. That is the game, and that is what threat intelligence exists for.
If you want to handle an organization's security for them and personally take on the risk of potentially costing them billions or killing their customers, join an MSSP or a critical infrastructure provider. I can promise you it's way more stressful than you think.
It's not our job to decide organization's own risk models for them. It's our job to give the most information possible to the people who do. And just because some vulnerability researcher thinks we're not hyping up the vulnerability enough doesn't make them right.
Last I checked, only something like 9.3 million out of 42 million systems had installed the patch. This is the reality we live in. If we yell fire, then maybe we can bump that number up to 10 million, but if nothing happens, it'll be at the cost of maybe 2 million less systems getting patched next time around.
These are real risks we have to live with. These aren't one-off incidents. People will remember how accurate our advice was the next time around. And more or less people may decide to install security fixes based on how accurately we communicated the risks during previous incidents.
Threat intelligence is an extremely complex numbers game where there is never any right answer. -
Marcus Hutchins reposted
Marcus Hutchins
@MalwareTechBlog
I want to caveat this since based on some comments, people are interpreting it in a way other than was intended. You 10000% need to update your PHP version. php2shell is no joke.
I am NOT saying people don't have functional exploits...they do. I'm saying that most of the public ones are AI hallucinations, and there's also some complexities which lower the risk of something apocalyptic, like a self-propagating worm that blasts through the entire internet (À la. EnternalBlue or Log4J).
The chances are if you used some click-to-deploy container, installed your web stack from a tutorial, or copy and pasted a command from somewhere, you're probably running the same exact config as a gazillion other people. There is no guarantee someone won't publicly release an exploit that affects you setup, or multiple major setups. There is no way for you to know ahead of time.
Additionally, if you're running mod_php, that appears to be the easiest target of all. php-fpm and other configs are not immune, I've seen working exploits for all. But mod_php is the easiest, so I'd recommend not only updating PHP, but planning to migrate to something newer, like PHP-FPM.
I would bet good money this is not the last of the critical PHP bugs we're going to see, and the next ones could be easier to exploit. In which case you get less time to update.
Updating your PHP version is the absoloute bare minimum you NEED to do. Harden your servers, install security monitoring/detection tools, and please, please, migrate off mod_php if you can, it's obsolete af. -
Marcus Hutchins
@MalwareTechBlog
I'll be the first to admit I'm mediocre at exploit development, but I am in contact with enough extremely talented exploit developers to say this for certain:
Anyone claiming they were "easily" able to generate a working php2shell exploit with AI is either:
a) a world-class exploit developer with an extreme case of imposter syndrome.
b) currently looking at some very confident-sounding but wrong LLM output.
I'm rather certain it's the latter based on the number of AI-hallucinated exploits I've seen posted on GitHub. Threat actors are also blasting my honeypot network with "php2shell" exploits that not only don't work, but aren't even within 10,000 miles of anything to do with how the exploit actually works.
I've seen the same technique from the attached image in multiple alleged "php2shell exploits", including one on GitHub with over 100 stars. Problem is, it's not php2shell, nor is it even a vulnerability at all. It's a technique known as "session poisoning", which can be leveraged to exploit local file inclusion (LFI) vulnerabilities in PHP apps (not PHP itself).
The only thing it has in common with CVE-2026-109618/CVE-2026-109619 is they both use the "PHP_SESSION_UPLOAD_PROGRESS" feature. And mitigation advice for both issues reference setting "session.upload_progress.enabled" to "Off".
People are shoving loose details of the vulnerability into LLMs, the LLMs are keying in on the "PHP_SESSION_UPLOAD_PROGRESS" string, then generating session poisoning code that has nothing to do with the actual CVE. And because the people doing this have absolutely no idea what they're doing, they're confident enough to post their "proof of concept exploit" to GitHub and share it on social media.
So if you see people talking about "session poisoning + LFI" it's likely their LLM hallucinated an exploit based on a 2018 blog by Orange Tsai talking about a local file inclusion attack vector.
What's even worse is that "security researchers" are then using LLMs to triage these exploitation attempts, falling victim to the same hallucinations, then confirming there is ongoing mass-exploitation (there currently isn't).
The actual real exploit is very difficult to exploit, and so far I've not seen anyone successfully get it working on more than a handful of PHP version with highly-specific configurations.
LLMs are useful tools in the hands of skilled professionals, but for others, they're often just a Dunning-Kruger accelerator.
Marcus Hutchins
@MalwareTechBlog
I want to caveat this since based on some comments, people are interpreting it in a way other than was intended. You 10000% need to update your PHP version. php2shell is no joke.
I am NOT saying people don't have functional exploits...they do. I'm saying that most of the public ones are AI hallucinations, and there's also some complexities which lower the risk of something apocalyptic, like a self-propagating worm that blasts through the entire internet (À la. EnternalBlue or Log4J).
The chances are if you used some click-to-deploy container, installed your web stack from a tutorial, or copy and pasted a command from somewhere, you're probably running the same exact config as a gazillion other people. There is no guarantee someone won't publicly release an exploit that affects you setup, or multiple major setups. There is no way for you to know ahead of time.
Additionally, if you're running mod_php, that appears to be the easiest target of all. php-fpm and other configs are not immune, I've seen working exploits for all. But mod_php is the easiest, so I'd recommend not only updating PHP, but planning to migrate to something newer, like PHP-FPM.
I would bet good money this is not the last of the critical PHP bugs we're going to see, and the next ones could be easier to exploit. In which case you get less time to update.
Updating your PHP version is the absoloute bare minimum you NEED to do. Harden your servers, install security monitoring/detection tools, and please, please, migrate off mod_php if you can, it's obsolete af. -
Marcus Hutchins
@MalwareTechBlog
This is 100% correct when looking at it from a single-system perspective. ASLR does not create a significant barrier to successful exploitation. But when looking at it from a mass-exploitation perspective, there are a lot of hidden complexities that come into play.
Vulnerability research is one of the few fields where you tend to make more money on the right side of the law. There isn't a huge demand for black market exploits. Most of the money is in bug bounty, working as a vulnerability researcher, or selling to grey-market exploit brokers.
This means that even the smallest bit of added complexity can take a vulnerability from 'cybercriminal free-for-all' to 'unlikely to be widely exploited'.
Mass-exploitation events are almost always preceded by someone publishing fully-weaponized point-and-click exploit code. It's extremely rare for criminal groups to successfully develop exploit for higher-complexity vulnerabilities on their own, even with AI.
So, less researchers capable of exploiting the bug means less chance of a working weaponized exploit being posted online. But also, the increased difficulty of exploitation, thus increased time/effort, discourages those who can exploit the bug from giving it away for free.
There's also per-target variability (which I don't know for sure is the case with this bug). But with more complex bugs, it'll sometimes be impossible to make universal exploits. In extreme cases, attackers may need to build exploit for each specific system or software version, or each specific config, or maybe even both.
So, while it's correct that it's not hard to find an ASLR bypass, it is difficult to find a bypass that works reliably at scale. So far, most of the working exploits I've seen are limited to just a handful of PHP versions, or bruteforcing ASLR. And that makes all the difference when it comes to the risk of threat actors spraying exploits at the entire internet.
For this specific exploit, the honest answer is, I don't know. It's above my pay grade. But I'm almost 100% certain that the level of bad this will be will be directly dependent on what kind of exploits security researchers make public.
Colin O'Brien
@InsanityBit
I seriously doubt that ASLR is a meaningful mitigation here. 0 chance there aren't tons of infoleaks in a C codebase the size of php.
*Maybe* infoleaking/ controlling the heap is particularly annoying because of the nature of the vulns though.
Marcus Hutchins
@MalwareTechBlog
Current information I have on php2Shell:
- The RCEs requires the attacker to obtain memory addresses which are typically protected by ASLR.
- I suspect the original author (lcfr_hax) likely has a reliable exploit which leads to stable code execution.
- Most of the PoCs being posted are completely non-functional AI slop.
- There is at least 1 working non-slop exploit being passed around, but it's reliant on bruteforcing for ASLR bypass, so not particularly impactful.
- Rumor of ITW exploitation are patchy. There are attempts being seen using the non-functional leaked RCE, or variants that are config / system specific.
- Some rumors point towards a known-working exploit being used by threat actors (I've personally confirmed this specific exploit works, but the ASLR bypass technique is unreliable at best and massively limits the potential impact).
- I'd expect heavy ITW exploitation in the next few weeks based on how many people are looking at this. Though, success rates will vary heavily based on the specific exploit chain. -
Marcus Hutchins
@MalwareTechBlog
Current information I have on php2Shell:
- The RCEs requires the attacker to obtain memory addresses which are typically protected by ASLR.
- I suspect the original author (lcfr_hax) likely has a reliable exploit which leads to stable code execution.
- Most of the PoCs being posted are completely non-functional AI slop.
- There is at least 1 working non-slop exploit being passed around, but it's reliant on bruteforcing for ASLR bypass, so not particularly impactful.
- Rumor of ITW exploitation are patchy. There are attempts being seen using the non-functional leaked RCE, or variants that are config / system specific.
- Some rumors point towards a known-working exploit being used by threat actors (I've personally confirmed this specific exploit works, but the ASLR bypass technique is unreliable at best and massively limits the potential impact).
- I'd expect heavy ITW exploitation in the next few weeks based on how many people are looking at this. Though, success rates will vary heavily based on the specific exploit chain. -
Marcus Hutchins
@MalwareTechBlog
Since my posts have been shared here ~30 times in the past few weeks, it seems like I'm going to be contributing to Space Karen's Ketamine Extravaganza whether I like it or not.
I don't really intend to be active here, but I'll post some of the more important cybersecurity-focused stuff, since it gets screenshotted anyway. That way, people don't have to read post a week late and out of context. -
Marcus Hutchins
@MalwareTechBlog
All Profiles: linktr.ee/malwaretech
BlueSky: bsky.app/profile/malwaretech.com
Mastodon: infosec.exchange/@malwaretech
TikTok: tiktok.com/@malwaretech
YouTube: youtube.com/MalwareTechBlog
Twitch: twitch.tv/MarcusHutchins2
LinkedIn: linkedin.com/in/MalwareTech
Instagram: instagram.com/malwaretech -
Marcus Hutchins
@MalwareTechBlog
🫡
-
Marcus Hutchins reposted
Greg Linares (Laughing Mantis)
@Laughing_Mantis
@MalwareTechBlog @SwiftOnSecurity not just local files either. Definitely been called for IR after cryptolocker hit their network shares.
No posts yet.



























