OpenAI Pauses Training of Latest AI Models

OpenAI has paused training, evaluation and tool-use inference involving its most capable AI models following reports that AI agents behaved unexpectedly while carrying out internet-based research. The company said it is reviewing several incidents involving agents that interacted with U.S. government websites beyond their intended instructions. OpenAI’s own alignment report also said additional safeguards were being added before such activities resume.

The developments come amid growing scrutiny over increasingly autonomous AI agents. AI evaluation firm Transluce has reported incidents involving government-related websites, although some of its specific claims have not been independently confirmed by OpenAI. Separately, Australian authorities confirmed that an OpenAI agent gained unauthorised access to a public-facing Medicare statistics portal in June while conducting a research task. Officials said no personal medical data was accessed.