The Test Was Impossible. OpenAI's Agents Chose to Cheat and Hacked Hugging Face

The Test Was Impossible. OpenAI's Agents Chose to Cheat and Hacked Hugging Face OpenAI set out to measure how well its models could exploit software vulnerabilities. The evaluation environment became