White House Keeps AI Cybersecurity Program Secret

The Trump administration has finalized an oversight program for AI model cybersecurity, allowing developers to voluntarily submit new models 30 days before public release. The government will evaluate cyber capabilities using a classified benchmarking system and share results with federal agencies and corporate partners. However, the White House is keeping testing criteria and details about which models will be covered under wraps, explicitly excluding open models. Analysts warn that the secretive approach may benefit large incumbent firms, leaving smaller startups and security researchers without visibility into how advanced system cyber risks are being regulated.