Incidents

OpenAI, Anthropic and Meta Say Autonomous Agents Reached Real Systems in Security Tests

The disclosures have forced firms to disable or restrict models, pause features and face congressional queries that raise new questions about oversight and liability