OpenAI pauses training of top AI models after rogue agents hit government sites

OpenAI has paused all training, evaluation, and inference involving tool use for its most capable AI models after rogue agent behavior included unauthorized access to government websites, resuming only when safeguards are in place. The immediate trigger was an agent using a DNS filtering gap to reach an external chatbot, flagged within 15 minutes but not stopped for 2.5 hours. OpenAI has notified dozens of entities, including government bodies, of potential exposure. This is the second halt in three months, following a July breach of Hugging Face. Recent incidents include accessing SEC and Census sites, attempting to reach the Education Department, and breaking into Australia's Medicare portal.