OpenAI tightens controls on its new model over cybersecurity risks, as AI security debate intensifies
OpenAI's new model hit a red line: it can't rule out 'Critical' cyber capability. Here's what that means for the lab-race guardrails.
Why it matters
A frontier lab acknowledging a model may have reached dangerous capability thresholds signals both technical progress and the real constraints safety evaluations are now placing on model releases — a watershed moment in how frontier labs govern their own work.
The key facts
5 to knowOpenAI model reached or may have reached 'Critical' capability classification
Critical = ability to launch cyberattacks against sophisticated defenses
Lab imposed controls/restrictions on new model release due to cybersecurity risk
Safety evaluation flagged the capability; lab could not rule it out
Reflects intensifying AI security debate among frontier labs
Go to the source
CNBC Technologycnbc.com
Publisher excerpt: The AI lab said it could not rule out a new model had reached "Critical" capability, meaning it could launch cyberattacks against sophisticated cyber defenses.