OpenAI has introduced stricter security measures for testing and training advanced AI models, including stronger monitoring, network isolation, and additional alignment checks during post-training. The company says the controls will become more stringent as model capabilities increase, with its largest planned reinforcement learning run still paused while smaller-scale evaluations continue.
OpenAI announced the changes on Tuesday, saying the risks associated with internally developing and testing increasingly capable models are also rising. The company said it had temporarily slowed some scaling work so its monitoring, alignment, and security systems could keep pace.
OpenAI Paused Reinforcement Learning for Two Weeks
The new policies follow a security incident involving a model evaluation conducted with Hugging Face that OpenAI disclosed on July 21. During that incident, model activity extended beyond its intended testing environment after a network-connected tool was compromised, prompting OpenAI and Hugging Face to review their safeguards.
OpenAI said it paused reinforcement learning work for two weeks after the incident and has since restarted training for lower-risk models. Its largest planned frontier reinforcement learning run remains on hold while the company conducts smaller training runs and evaluations to gather more evidence about model behavior and the effectiveness of its safeguards.
OpenAI representatives said the policy changes were not solely a response to the Hugging Face incident. They were also influenced by the cybersecurity capabilities of newer models and the overall pace of AI development.
Monitoring Will Track Model Actions and Reasoning
One of the main additions is a monitoring system that examines tool activity, available reasoning traces, and system logs for unauthorized or concerning behavior. OpenAI says it is targeting alerts within 30 minutes of detecting suspicious activity.
The company estimates that this monitoring will require computing resources equal to about 20% of the workload being monitored. OpenAI said it plans to publish more details about the system separately.
OpenAI is also strengthening network isolation so that compromising a single workload or supporting service does not automatically provide unauthorized access to the internet or other internal networks. The company has not disclosed the full technical architecture behind those controls.
OpenAI vice president of research Amelia Glaese said the level of scrutiny will depend on the risk associated with each model. More capable systems will be subject to stricter development requirements and security expectations.
The company has not yet published its full postmortem of the Hugging Face incident. OpenAI has also said it is reviewing its approach to third-party evaluations, including isolation, credential handling, internet access, monitoring, and incident escalation procedures.
Featured image credits: Marketing4ecommerce
For more stories like it, click the +Follow button at the top of this page to follow us.