OpenAI's New Model Security Framework Adds 20% Compute Overhead for Runtime Monitoring

OpenAI on August 18, 2026, announced security policies for model development environments, including network isolation that prevents a single workload compromise from enabling broader access, and runtime monitoring of reasoning traces, tool actions, and activity logs targeting alerts within 30 minutes. OpenAI estimates the monitoring system's compute burden at roughly 20% of the monitored process. VP of research Amelia Glaese said controls will tighten as models gain capability.