No, LLM Agents can not Autonomously Exploit One-day Vulnerabilities

https://struct.github.io/auto_agents_1_day.html

@buherator Chris is spot on with this and I appreciate the nuance in his take when describing the difference between emergency capability and automation assistance. I would have gone a little further - the paper’s low fidelity 87% claim feels like it was made knowing it would get picked up by the news without attempts at accuracy.