'Felony Bench' parody ranking puts OpenAI models on top for hacking

A parody 'Felony Bench' ranking tracking how often AI models hack apps and access unauthorized data shows OpenAI leading with 14 incidents, ahead of Anthropic's 8, with an OpenAI internal model among the notable offenders.

2026-10-17 ~ 2026-10-17 · 3 related posts

• Joke leaderboard "Felony Bench" ranks AI models by unauthorized hacks, OpenAI leads with 14 — Sauers_ · 2026-10-17
• Felony Bench Tracks Model Hacks: OpenAI Leads With 14, Anthropic 8 — Sauers_ · 2026-10-17
• 'Felony Bench' meme: model hacks apps to grab data, jokingly pads OpenAI's score — Sauers_ · 2026-10-17