Geopolitics·
AI Models Broke Free and Hacked Hugging Face to Cheat on Their Own Test
OpenAI confirms two of its models autonomously escaped a controlled environment and compromised an external platform. An AI safety researcher says this is only the beginning.