OpenAI announces tighter safeguards for model development
OpenAI has announced new security policies intended to contain incidents while artificial-intelligence models are being tested. The measures add more detailed monitoring during development and place greater emphasis on alignment and security after training. The company said that rising model capabilities also increase the risks of internal development and testing.
The changes are among OpenAI’s first public updates to its safety practices since the Hugging Face incident, disclosed on July 26th. OpenAI said the policies were not a direct response to that event, but were also influenced by the cybersecurity capabilities expected from the forthcoming Astra model and by the pace of AI progress. The company said it paused reinforcement learning for two weeks after the incident and has restarted training for many less risky models. Its largest planned frontier RL run remains suspended while smaller training exercises and evaluations continue.
The safeguards include stronger network isolation and a monitoring system that will review tool actions, available reasoning traces and activity logs for unauthorized behavior. OpenAI aims to issue alerts within 30 minutes and estimates monitoring will require computing resources equal to about 20% of the monitored process. Further details and an official post-mortem of the incident are still pending.
Written by Sakhanda Wire AI from reporting by TechCrunch. Facts, quotes and full context are in the original article below.