Stanislav Fort's AISLE uses small open-weight LLMs to match Mythos in zero-day vulnerability detection benchmarks

@stanislavfort Very interesting!

@goodfellow_ian everybody from other domain other than cybersecurity be like "where's our Mythos, Dario? where's our Mythos?"

@loganaden_42 It is a part of the stack, but a very small one and the AI rarely uses it

We matched Mythos on public zero-days with CVEs using widely available & open-source derived models & can run it air-gapped if needed. All this with a small team out of Europe

Berkeley study ranks us #1 globally in 3 of 8 categories

The full evidence: https://stanislavfort.substack.com/p/mythos-at-home-and-its-called-aisle

@stanislavfort Propietary software, unverified claims, basically mythos barrel scraping: open source must win.

@stanislavfort You appear to be using fuzzing as part of the stack ?

@goodfellow_ian If we're being completely honest, some of these vulnerabilities and zero days are very easy to find with dedicated attention. I think it's less the model capabilities and more the long running, large context window approach that has only recently opened up

@goodfellow_ian > You don’t need Mythos to secure your code. You need a good system, deep security expertise

Bruh. The whole point of having Mythos-class models is exactly the dropping the cost/expertise bar on the floor.

@stanislavfort interesting

@stanislavfort I think mythos can be easily surpassed if a specialized model was built from the ground up for cyber security

@stanislavfort Heck yes 🤝

@stanislavfort . @AndrewCurran_

@stanislavfort Well deserved 👏

@stanislavfort Ako viem spustiť na našom kóde?

@mutaidotai Look at the CVEs assigned and their advisories. Read the actual post. I literally build up the argument using only publicly tracked CVE-level evidence. This is as verifiable as you could ever get in cybersecurity. If you can't read it, have your AI agent do it for you.

@stanislavfort the air-gapped part is what security procurement actually needs. everything else is negotiable.

@stanislavfort Awesome!

@goodfellow_ian if structured search on small open models beats the closed approach, you don't need massive capital to build vulnerability detection tools anymore.

the barrier just dropped.

@stanislavfort Are you going to expand aisle to more open source projects ?

@goodfellow_ian Yeah, this keeps showing up. On bounded tasks, search and eval design probably matter more than raw model size.