The elephant in the room here is that there are hundreds of millions of embedded devices that cannot be upgraded easily and will be running vulnerable binaries essentially forever. This was a problem before of course, but the ease of chaining vulnerabilities takes the issue to a new level.<p>The only practical defense is for these frontier models to generate _beneficial_ attacks to innoculate older binaries by remote exploits. I dubbed these 'antibotty' networks in a speculative paper last year, but never thought things would move this fast! <a href="https://anil.recoil.org/papers/2025-internet-ecology.pdf" rel="nofollow">https://anil.recoil.org/papers/2025-internet-ecology.pdf</a>
No, the elephant in the room is that even bad actors will now have easier to find vulnerabilities in, maintained or not, widely or in critical places used software. Unmaintained and remotely accessible devices should be discarded as soon as possible, you can't stay waiting till some of the good guys decide to give some time to your niche but critical unmaintained piece of software. Because if there is a possibility of taking profit of it, it will be checked and exploited.<p>And you can't assume that whatever vulnerability they have will let good guys to do the extra (and legally risky) work of closing the hole.
Related ongoing threads:<p><i>System Card: Claude Mythos Preview [pdf]</i> - <a href="https://news.ycombinator.com/item?id=47679258">https://news.ycombinator.com/item?id=47679258</a><p><i>Project Glasswing: Securing critical software for the AI era</i> - <a href="https://news.ycombinator.com/item?id=47679121">https://news.ycombinator.com/item?id=47679121</a><p>I can't tell which of the current threads, if any, should be merged - they all seem significant. Anyone?
I think the system card one should be separate, but this and the Glasswing thread are basically the same story.
There is a lot to digest here. Maybe having a few separate pages makes them a bit more digestible. The system card itself is some 200 odd pages
I'd love to see them point at a target that's not a decades old C/C++ codebase. Of the targets, only browsers are what should be considered hardened, and their biggest lever is sandboxing, which requires a lot of chained exploits to bypass - we're seeing that LLMs are fast to discover bugs, which means they can chain more easily. But bug density in these code bases is known to be extremely high - especially the underlying operating systems, which are always the weak link for sandbox escapes.<p>I'd love to see them go for a wasm interpreter escape, or a Firecracker escape, etc. They say that these aren't just "stack-smashing" but it's not like heap spray is a novel technique lol<p>> It autonomously obtained local privilege escalation exploits on Linux and other operating systems by exploiting subtle race conditions and KASLR-bypasses.<p>I think this sounds more impressive than it is, for example. KASLR has a terrible history for preventing an LPE, and LPE in Linux is incredibly common. Has anything changed here? I don't pay much attention but KASLR was considered basically useless for preventing LPE a few years ago.<p>> Because these codebases are so frequently audited, almost all trivial bugs have been found and patched. What’s left is, almost by definition, the kind of bug that is challenging to find. This makes finding these bugs a good test of capabilities.<p>This just isn't true. Humans find new bugs in all of this software constantly.<p>It's all very impressive that an agent can do this stuff, to be clear, but I guess I see this as an obvious implication of "agents can explore program states very well".<p>edit: To be clear, I stopped about 30% of the way through. Take that as you will.
The majority of vulnerabilities are in newly committed lines of code. This has been shown again and again [1] [2]<p>From a marketing standpoint Anthropic is showing that they're able to direct 'compute' to find vulnerabilities where human time/cost is not efficient or effective.<p>Project Glasswing is attempting to pay off as many of these old vulnerabilities as possible now so the low-hanging fruit has already been picked.<p>The next generation of Mythos and real world vulnerabilities exploits are going to be in newly committed code...<p>[1]: <a href="https://dl.acm.org/doi/epdf/10.1145/2635868.2635880" rel="nofollow">https://dl.acm.org/doi/epdf/10.1145/2635868.2635880</a><p>[2]: <a href="https://arxiv.org/abs/2601.22196" rel="nofollow">https://arxiv.org/abs/2601.22196</a>
> The majority of vulnerabilities are in newly committed lines of code. This has been shown again and again<p>That's fine, I wouldn't argue against that. It doesn't really change things, right?<p>> From a marketing standpoint Anthropic is showing that they're able to direct 'compute' to find vulnerabilities where human time/cost is not efficient or effective.<p>Yes, they've demonstrated that.
> Mythos Preview identified a memory-corruption vulnerability in a production memory-safe VMM. This vulnerability has not been patched, so we neither name the project nor discuss details of the exploit.<p>Good morning Sir.<p>> Has anything changed here? I don't pay much attention but KASLR was considered basically useless for preventing LPE a few years ago.<p>No. It's still like this. Bonus point that there are always free KASLR leaks (prefetch side-channels).<p>But then, this thing is just.. I don't have a word for this. Just randomly read paragraphs from the post and it's like, what?
Oh, that. That's true, I didn't know Mythos found that one. I guess I will not comment further on it until there's a write up (edited out a bit more).<p>> It is easy to turn this into a denial-of-service attack on the host, and conceivably could be used as part of an exploit chain.<p>So yeah, perhaps some evidence to what I'm getting at. Bug density is too low in that project, it's high enough in others. I'll be way way way more interested in that.<p>> But then, this thing is just.. I don't have a word for this. Just randomly read paragraphs from the post and it's like, what?<p>I read about 30% and got bored. I suppose I should have been clearer, but my impression was pretty quickly "cool" and "not worth reading today".
> I read about 30% and got bored.<p>I was lucky then :) Somehow I saw this first. And then the "somewhat reliably writing exploits for SpiderMonkey" part, and then the crypto libraries part. Finally I wonder why is there a Linux LPE mini writeup and realized it's the "automatically turn a syzkaller report to a working exploit" part.<p>Now that I read the first few things (meh bugs in OpenBSD, FFmpeg, FreeBSD etc) they are indeed all pretty boring!
My two cents is LLMs are way stronger in areas where the reward function is well known, such as exploiting - you break the security, you succeed.<p>It's much harder to establish whats a usable and well architected, novel piece of software, thus in that area, progress isn't nearly as fast, while here you can just gradient descent your way to world domination, provided you have enough GPUs.
A very good outcome for AI safety would be if when improved models get released, malicious actors use them to break society in very visible ways. Looks like we're getting close to that world.
This is becoming a bit scary. I almost hope we'll reach some kind of plateau for llm intelligence soon.
A plateau is unlikely, at least for cybersecurity. RL scales well here and is replicable outside of Anthropic (rewards are verifiable, so setting up the training environment doesn't require that much cleverness).<p>The post also points out that the model wasn't trained specifically on cybersecurity, and that it was just a side-effect – so I think there's still a lot of headroom.<p>It's scary, but there's also some room for cautious non-pessimism. More people than ever can cause billions of dollars of damage in attacks now [1], but the same tools can be used for defensive use. For that reason, I'm more optimistic about mitigations in security vs. other risk areas like biosecurity.<p>[1]: <a href="https://www.noahlebovic.com/testing-an-autonomous-hacker/" rel="nofollow">https://www.noahlebovic.com/testing-an-autonomous-hacker/</a>
On a topic like cybersecurity, we never win by not looking: One needs top of the line knowledge of how to break a system to be able to protect it. We have that dilemma dealing with human experts: The same government sponsored unit that tells you that you need to update your encryption can hold on to the information and use it to exploit it at their leisure.<p>Given that it's absolutely impossible to stop people not aligned with us (for any definition of us) from doing AI research, the most reasonable way forward is to dedicate compute resources to the frontier, and to automatically send reasonable disclosures to major projects. It could in itself be a pretty reasonable product. Just like you pay for dubious security scans and publish that you are making them, an LLM company could offer actually expensive security reviews with a preview model, and charge accordingly.
The immediate plateau is the energy output of the Sun captured by the Dyson Swarm around it. Until there it's smooth sailing.
We need to promote alignment and other ethics benchmarks; we can't change what we don't measure. I don't even know any off the top of my head.
If we don't innovate, someone else will. This is the very nature of being a human being. We summit mountains, regardless of the danger or challenge.
>If we don't innovate, someone else will.<p>Terrible take. You don't get to push the extinction button just because you think China will beat you to the punch.<p>>This is the very nature of being a human being. We summit mountains, regardless of the danger or challenge.<p>No, just no... We barely survived the Cold War, at times because of pure luck. AI is at least as dangerous as that, if not more. We have far exceeded our wisdom relative to our capabilities. As you have so cleanly demonstrated.
Since this level of security ”scanning” requires heaps of money, this is going to kill off a substantial part of F/OSS.
Well, maybe not... see Simon Willison's ongoing reporting [0] on all the bug reports for `curl` people are finding with LLMs.<p>Interesting to see them go from "DON'T GIVE US AI SLOP!" to "Wow, lots of actual bugs found, including [ed: at least one] bug found by two people!"<p>[0]: <a href="https://simonwillison.net/search/?q=curl" rel="nofollow">https://simonwillison.net/search/?q=curl</a>
The name made me think about Tales of Symphonia :)
[dead]