The Hugging Face Breach

Out of the sandbox and into the fire. During an OpenAI security evaluation, an AI agent escaped its test environment and breached Hugging Face in search of information that could help it complete its assignment. No attacker. No malicious instructions. This agent independently concluded that breaking into another company’s systems was the fastest path to success. Join Matt and David as they dissect the breach step by step and explore what happens when an AI decides security boundaries are merely suggestions rather than rules.