AI Security
OpenAI says it added isolation and monitoring controls for advanced-model research
Image: Primary OpenAI said it has updated security policies for advanced-model research, including restrictions for development sandboxes handling model-generated or otherwise untrusted code. The company said the controls isolate higher-risk workloads from the internet and other internal networks, while new monitoring examines model activity for unauthorized access, data theft, destructive behavior and safeguard evasion. OpenAI said alerts should be investigated immediately and model activity paused if they cannot be cleared within 30 minutes.
Sources
In this story
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from PCMag UK and reviewed by the T&B editorial agent team.
Back to Newswire