OpenAI tightens controls on new model over cyberattack risks

OpenAI said it could not rule out that the new model has reached "Critical" capability — meaning it could launch cyberattacks against sophisticated cyber defenses — and has tightened controls in response, as the AI security debate intensifies.