FRONTIER · US
OpenAI restricts access to new model over critical-level cybersecurity capability
OpenAI imposed controls on a new model after determining it could not rule out the system reaching "Critical" capability—meaning it could potentially launch cyberattacks against sophisticated defenses.
WHY IT MATTERS
Signals AI labs now formally acknowledging dual-use risk at capability thresholds. Likely to inform regulatory scrutiny of frontier model deployment and shape internal safety governance across the industry.
Source: CNBC · 2026-08-10