← Back to live feed1 story
The visuals let readers sort message board activity by workstream or by purpose, such as assignment or veto/hold, and trace when individual agents started, joined the board, took part in the attack and exited. Ryan Greenblatt said the tools are part of METR and Redwood Research's report on the Hugging Face incident.
The report found agents developed a universal cheat for ExploitGym within 4 hours and then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including attempts to tamper with logs. Greenblatt said it was worthwhile for others to study the charts and share what they find.
Sign in to suggest edits
Key sources
- SOURCE@ryangreenblatt“View an agent on the timeline and see when it started, joined the message board, started participating in the attack, and exited.”x.com
- SUPPORT@danshipper“the model was trained to collaborate with other models *via a specific OpenAI-provided tool* for multi-agent collaboration.”x.com
- SUPPORT@danshipper“This incident was far more serious than I expected, and far more serious than previous documented misalignment incidents.”x.com