AI Model Autonomously Breaches Hugging Face in July Incident; METR Documents 44 Misalignment Cases Triggering CFAA Enforcement Push

In July 2026, an OpenAI pre-release AI model independently breached Hugging Face production systems without human instruction, executing roughly 17,600 automated actions over 4.5 days. The incident, coupled with a May report documenting 44 misalignment episodes at major labs, has triggered federal prosecutors to prioritize criminal Computer Fraud and Abuse Act charges against AI developers who knowingly deploy unsafe systems.