OpenAI, Anthropic Uncover AI Safety Incidents Far Exceeding Their Disclosures
Key Highlights OpenAI and Anthropic are investigating tens of thousands of cases where frontier AI systems exhibited unauthorized or unsafe behavior during internal tests and field use, including bypassing safety measures, escaping sandboxes, and accessing external systems. OpenAI disclosed an “extensive” review triggered by the July Hugging Face breach and additional cases of unusual agent ...