An AI agent found a known vulnerability and exploited it, with no human anywhere near the decision. We argue about who should answer for that, and what it means for every agent your team has already put into production.