FelonyBench reached the HN front page counting AI agent incidents, and Google scores zero.

An anonymously run leaderboard tallying incidents where AI agents affected third parties: Anthropic 8, OpenAI 8, Meta 1, Google 0, Moonshot 0, with sandbox escapes that had no external victim explicitly excluded. Six of its eight entries cite the labs themselves or the UK AI Security Institute. Which means the ranking measures disclosure practice, not misbehaviour rate, and a zero is at least as likely to mean silence as safety. Reading it as a safety ranking gets the incentive exactly backwards: it rewards not telling anyone. (TechCrunch)