It's a myth that you need Mythos to find bugs: Open source models can do it just as well

At Black Hat Asia, Ari Herbert-Voss (RunSybil CEO and OpenAI's first security hire) argued that open-source LLMs can match Anthropic's Mythos for automated bug-finding by using "scaffolding" to run multiple models together; he cited supralinear scaling as a driver of capability, noted Mythos' restricted access and cost, and emphasized that human experts remain necessary to orchestrate models and validate the high volume of findings (similar to fuzzing), which will likely drive broader AI adoption in infosec.