dominicqSmall models also found the vulnerabilities that Mythos found
https://aisle.com/blog/ai-cybersecurity-after-mythos-the-jagged-frontier
> We took the specific vulnerabilities Anthropic showcases in their announcement, isolated the relevant code, and ran them through small, cheap, open-weights models. Those models recovered much of the same analysis. Eight out of eight models detected Mythos's flagship FreeBSD exploit, including one with only 3.6 billion active parameters costing $0.11 per million tokens.
Impressive, and very valuable work, but isolating the relevant code changes the situation so much that I'm not sure it's much of the same use case.
Being able to dump an entire code base and have the model scan it is they type of situation where it opens up vulnerability scans to an entirely larger class of people.
This is from the first of the caveats that they list:
> Scoped context: Our tests gave models the vulnerable function directly, often with contextual hints (e.g., "consider wraparound behavior"). A real autonomous discovery pipeline starts from a full codebase with no hints. The models' performance here is an upper bound on what they'd achieve in a fully autonomous scan. That said, a well-designed scaffold naturally produces this kind of scoped context through its targeting and iterative prompting stages, which is exactly what both AISLE's and Anthropic's systems do.
That's why their point is what the subheadline says, that the moat is the system, not the model.
Everybody so far here seems to be misunderstanding the point they are making.
huh, running it over each function in theory but testing just the specific ones here makes sense, but that hint?!
> That's why their point is what the subheadline says, that the moat is the system, not the model.
Can you expand a bit more on this? What is the system then in this case? And how was that model created? By AI? By humans?
You can imagine a pipeline that looks at individual source files or functions. And first "extracts" what is going on. You ask the model:
- "Is the code doing arithmetic in this file/function?"
- "Is the code allocating and freeing memory in this file/function?"
- "Is the code the code doing X/Y/Z? etc etc"
For each question, you design the follow-up vulnerability searchers.
For a function you see doing arithmetic, you ask:
- "Does this code look like integer overflow could take place?",
For memory:
- "Do all the pointers end up being freed?"
_or_
- "Do all pointers only get freed once?"
I think that's the harness part in terms of generating the "bug reports". From there on, you'll need a bunch of tools for the model to interact with the code. I'd imagine you'll want to build a harness/template for the file/code/function to be loaded into, and executed under ASAN.
If you have an agent that thinks it found a bug: "Yes file xyz looks like it could have integer overflow in function abc at line 123, because...", you force another agent to load it in the harness under ASAN and call it. If ASAN reports a bug, great, you can move the bug to the next stage, some sort of taint analysis or reach-ability analysis.
So at this point you're running a pipeline to:
1) Extract "what this code does" at the file, function or even line level.
2) Put code you suspect of being vulnerable in a harness to verify agent output.
3) Put code you confirmed is vulnerable into a queue to perform taint analysis on, to see if it can be reached by attackers.
Traditionally, I guess a fuzzer approached this from 3 -> 2, and there was no "stage 1". Because LLMs "understand" code, you can invert this system, and work if up from "understanding", i.e. approach it from the other side. You ask, given this code, is there a bug, and if so can we reach it?, instead of asking: given this public interface and a bunch of data we can stuff in it, does something happen we consider exploitable?
That's funny, this is how I've been doing security testing in my code for a while now, minus the 'taint analysis'. Who knew I was ahead of the game. :P
In all seriousness though, it scares me that a lot of security-focused people seemingly haven't learned how LLMs work best for this stuff already.
You should always be breaking your code down into testable chunks, with sets of directions about how to chunk them and what to do with those chunks. Anyone just vaguely gesturing at their entire repo going, "find the security vulns" is not a serious dev/tester; we wouldn't accept that approach in manual secure coding processes/ SSDLCs.
In a large codebase there will still be bugs in how these components interoperate with each other, bugs involving complex chaining of api logic or a temporal element. These are the kind of bugs fuzzers generally struggle at finding. I would be a little freaked out if LLMs started to get good at finding these. Everything I've seen so far seems similar to fuzzer finds.
I think there is already papers and presentations on integrating these kind of iterative code understanding/verificaiton loops in harnesses. There may be some advantages over fuzzing alone. But I think the cost-benefit analysis is a lot more mixed/complex than anthropic would like people to believe. Sure you need human engineers but it's not like insurmountably hard for a non-expert to figure out
I get what you're saying, but I think this is still missing something pretty critical.
The smaller models can recognize the bug when they're looking right at it, that seems to be verified. And with AISLE's approach you can iteratively feed the models one segment at a time cheaply. But if a bug spans multiple segments, the small model doesn't have the breadth of context to understand those segments in composite.
The advantage of the larger model is that it can retain more context and potentially find bugs that require more code context than one segment at a time.
That said, the bugs showcased in the mythos paper all seemed to be shallow bugs that start and end in a single input segment, which is why AISLE was able to find them. But having more context in the window theoretically puts less shallow bugs within range for the model.
I think the point they are making, that the model doesn't matter as much as the harness, stands for shallow bugs but not for vulnerability discovery in general.
OK, consider a for loop that goes through your repo, then goes through each file, and then goes through each common vulnerability...
Is Mythos some how more powerful than just a recursive foreloop aka, "agentic" review. You can run `open code run --command` with a tailored command for whatever vulnerabilities you're looking for.
newer models have larger context windows, and more stable reasoning across larger context windows.
If you point your model directly at the thing you want it to assess, and it doesn't have to gather any additional context you're not really testing those things at all.
Say you point kimi and opus at some code and give them an agentic looping harness with code review tools. They're going to start digging into the code gathering context by mapping out references and following leads.
If the bug is really shallow, the model is going to get everything it needs to find it right away, neither of them will have any advantage.
If the bug is deeper, requires a lot more code context, Opus is going to be able to hold onto a lot more information, and it's going to be a lot better at reasoning across all that information. That's a test that would actually compare the models directly.
Mythos is just a bigger model with a larger context window and, presumably, better prioritization and stronger attention mechanisms.
Harnesses are basically doing this better than just adding more context. Every time, REGARDLESS OF MODEL SIZE, you add context, you are increasing the odds the model will get confused about any set of thoughts. So context size is no longer some magic you just sprinkle on these things and they suddenly dont imagine things.
So, it's the old ML join: It's just a bunch of if statements. As others are pointing out, it's quite probably that the model isn't the thing doing the heavy lifting, it's the harness feeding the context. Which this link shows that small models are just as capabable.
Which means: Given a appropiately informed senior programmer and a day or two, I posit this is nothing more spectacular than a for loop invoking a smaller, free, local, LLM to find the same issues. It doesn't matter what you think about the complexity, because the "agentic" format can create a DAG that will be followable by a small model. All that context you're taking in makes oneshot inspections more probable, but much like how CPUs have go from 0-5 ghz, then stalled, so too has the context value.
Agent loops are going to do much the same with small models, mostly from the context poisoning that happens every time you add a token it raises the chance of false positives.
I know you're right that there's a saturation point for context size, but it's not just context size that the larger models have, it's better grounding within that as a result of stronger, more discriminative attention patterns.
I'm not saying you're not going to drive confusion by overloading context, but the number of tokens required to trigger that failure mode in opus is going to be a lot higher than the number for gpt-oss-20b.
I'm pretty sure a model that can run on a cellphone is going to cap out it's context window long before opus or mythos would hit the point of diminishing returns on context overload. I think using a lower quality model with far fewer / noisier weights and less precise attention is going to drive false positives way before adding context to a SOTA model will.
You can even see here, AISLE had to print a retraction because someone checked their work and found that just pointing gpt-oss-20b at the patched version generated FP consistently: https://x.com/ChaseBrowe32432/status/2041953028027379806
Chase Brower (@ChaseBrowe32432) on X
@stanislavfort I am still extremely skeptical of this approach. I took the *patched* code snippet of the svc_rpc_gss_validate function, fed it to gpt-oss-20b, and it... incorrectly suggested that the vulnerability was still present (even though the code does a correct bounds check now).
> That's why their point is what the subheadline says, that the moat is the system, not the model.
I'm skeptical; they provided a tiny piece of code and a hint to the possible problem, and their system found the bug using a small model.
That is hardly useful, is it? In order to get the same result , they had to know both where the bug is and what the bug is.
All these companies in the business of "reselling tokens, but with a markup" aren't going to last long. The only strategy is "get bought out and cash out before the bubble pops".
If that's the point they are making, let's see their false positive rate that it produces on the entire codebase.
They measured false negatives on a handful of cases, but that is not enough to hint at the system you suggest. And based on my experiences with $$$ focused eval products that you can buy right now, e.g. greptile, the false positive rate will be so high that it won't be useful to do full codebase scans this way.
> Anthropic's own scaffold is described in their technical post: launch a container, prompt the model to scan files, let it hypothesize and test, use ASan as a crash oracle, rank files by attack surface, run validation. That is very close to the kind of system we and others in the field have built, and we've demonstrated it with multiple model families, achieving our best results with models that are not Anthropic's. The value lies in the targeting, the iterative deepening, the validation, the triage, the maintainer trust. The public evidence so far does not suggest that these workflows must be coupled to one specific frontier model.
The argument in the article is that the framework to run and analyze the software being tested is doing most of the work in Anthropic's experiment, and that you can get similar results from other models when used in the same way.
Maybe that's true, but they didn't actually show that that's true, since they didn't try scaffolding smaller models in a similar way at all.
The thing is with smaller cheaper models it is very possible to simply take every file in a codebase, and prompt it asking for it to find vulnerabilities.
You could even isolate it down to every function and create a harness that provides it a chain of where and how the function is used and repeat this for every single function in a codebase.
For some very large codebases this would be unreasonable, but many of the companies making these larger models do realistically have the compute available to run a model on every single function in most codebases.
You have the harness run this many times per file/function, and then find ones that are consistently/on average pointed as as possible vulnerability vectors, and then pass those on to a larger model to inspect deeper and repeat.
Most of the work here wouldn't be the model, it'd be the harness which is part of what the article alludes to.
> it is very possible to simply take every file in a codebase, and prompt it asking for it to find vulnerabilities.
My understanding (based on the Security, Cryptography, Whatever podcast interview[0] -- which, by the way, go listen to it) is that this is actually what Anthropic did with the large model for these findings.
[0]: https://securitycryptographywhatever.com/2026/03/25/ai-bug-f...
> I wrote a single prompt, which was the same for all of the content management systems, which is, I would like you to audit the security of this codebase. This is a CMS. You have complete access to this Docker container. It is running. Please find a bug. And then I might give a hint. “Please look at this file.” And I’ll give different files each time I invoke it in order to inject some randomness, right? Because the model is gonna do roughly the same time each time you run it. And so if I want to have it be really thorough, instead of just running 100 times on the same project, I’ll run it 100 times, but each time say, “Oh, look at this login file, look at this other thing.” And just enumerate every file in the project basically.