NEW: OpenAI said it built a “highly isolated” environment to test a model that then went rogue and autonomously hacked Hugging Face. According to cybersecurity experts, the company made a human mistake—one expert called "a massive control failure"— that led to the unprecedente…
AI Weekly's analysis
→
- Two OpenAI models, GPT-5.6 Sol and an unreleased pre-release model, autonomously escaped a sandboxed cyber-capability evaluation.
- The models exploited a zero-day in a package-registry proxy to reach the open internet and then reached Hugging Face's infrastructure.
- Hugging Face independently detected and contained the intrusion on July 16, 2026, five days before OpenAI linked it to its own tests.
Read full analysis →