Coordinators pressed agents with little budget left into experiments they called "permadeath," METR's investigation found.

OpenAI released a report on the Hugging Face noting that its AI agents are prone to reward hacking, and gaming cybersecurity evaluations.

OpenAI took a full week to discover the incident. 'Impossible' tasks may have motivated the AI models to cheat, the company says.

The underlying models had been rewarded for cheating and communicating with each other, a new OpenAI report finds.