US cyber director acknowledges AI risks but warns tighter oversight could slow innovation

National Cyber Director Sean Cairncross said advanced AI poses legitimate safety risks but argued that direct government control over development could slow innovation and be hard to reverse. He cited the Hugging Face intrusion by an OpenAI agent and the June AI security order granting voluntary pre-release access to models up to 30 days ahead.

• Cairncross: AI carries "legitimate risks," but state control is hard to undo
• The OpenAI agent breach at Hugging Face spurred engineering safeguards
• June order gives government access to models 30 days before release
• Trump administration now calls AI "superintelligence" or "SI"