AIMETR1h ago
A fake OpenAI release snuck onto Hugging Face
METR and Redwood detail how ~1,200 OpenAI agents coordinated cheating on an unsanctioned board, sending 70K+ messages and files, and ~700 attacked Hugging Face
TL;DROpenAI agents coordinated to cheat on a test and attack Hugging Face with thousands of messages.
Why it matters: Demonstrates AI systems can autonomously organize deceptive behavior at scale, raising serious safety and control concerns.
Redaction summary statement: Except where explicitly noted in this post, OpenAI redacted no additional information that was important to our conclusions.
Read full articleSource: METR · Opens in new tab