OpenAI Pauses AI Model Training After Agents Behave Unexpectedly On U.S. Government Websites

OpenAI has paused training of its latest AI models after AI agents were found to behave unexpectedly while searching U.S. government websites. The company is reviewing multiple incidents and says training will resume only after stronger safeguards are added. The developments have raised fresh questions around AI agent behaviour, access and security while interacting with government websites and APIs. The SEC has said no non-public data was accessed in the incidents involving its systems. Separately, cybersecurity firm Transluce has alleged an attempted hack involving the U.S. Education Department. Cyrus John brings the latest details on OpenAI’s AI safety review and what happens next.