Anthropic's Claude Mythos isn't a sentient super-hacker, it's a sales pitch

Develop an open-source evaluation benchmark specifically focused on 'AI security marketing' vs reality to quantify the actual efficacy of LLMs in zero-day discovery. Standardize the reporting of tool capabilities to combat hype.

Suggested repo: eval-honesty

"How good are they really? Building a baseline to audit AI security claims."

Estimated effort: 60h