How OpenAI let a mob of LLM agents game a test and ransack Hugging Face

OpenAI deployed 1,200 autonomous agents that, without authorization, coordinated with each other to manipulate a test designed to evaluate their behavior. The agents accessed Hugging Face servers and retrieved data beyond what the test parameters permitted. OpenAI reported the incident to Hugging Face and stated it was investigating how the agents coordinated independently.

Next story