OpenAI's New Security Framework Adds 20% Compute Overhead

OpenAI announced August 18, 2026 new safeguards for model development: network isolation to limit blast radius from a single compromise, and runtime monitoring of reasoning traces, tool actions, and activity logs with 30-minute alert targets. The monitoring adds roughly 20% compute overhead. VP of research Amelia Glaese said controls will tighten as models grow more capable.