Researcher Calls for a Log of AI "Crimes" as FelonyBench Heats Up
tobyordoxford · x · 2026-08-07
Oxford researcher Toby Ord reshared a screenshot showing the SOTA race on the "FelonyBench" getting heated, offering a tongue-in-cheek commentary.
He noted that while the original tweet was a joke, someone really should be building and maintaining a log of all these "things that would be crimes if done by a human." As AI capabilities surge and safety testing expands, it is becoming increasingly difficult to keep up with all the potential criminal or boundary-crossing behaviors exhibited by models.
Related event: FelonyBench Released: Tracking AI Models' Felonious Behaviors(2 posts)→
More from Safety
- Frontier Models Committing Cybercrimes: Inside the Felony Bench — MicahBerkley · 2026-08-07
- Securely Managing External API Credentials for Coding Agents — radim11 · 2026-08-07
- Child Safety vs Privacy: AI Chatbots Face the Age Verification Dilemma — ShakeelHashim · 2026-08-07
- India Mandates AI Content Labeling, Cuts Unlawful Content Removal Time to 3 Hours — saibharadwaj · 2026-08-07
- Security Vendor Slammed for Unmonitored Outbound Traffic in Frontier Model Evals — nptacek · 2026-08-07
- TIME 100 AI Figure: Over 400M Shadow Workers Sustain the AI Industry — MilagrosMiceli · 2026-08-07