The Trump administration has cooked up a cyber security plan for advanced AI models but, like a secret agent, it’s keeping details under wraps. Officials invited top AI companies to briefings at the White House, where they outlined an oversight framework that allows voluntary model submissions and vetting by federal agencies.

Smaller startups and safety advocates are left in the cold, wondering what's under the hood of this new rulebook. They argue transparency is key: 'This is not a handshake deal with tech companies,' says Brad Carson of Americans for Responsible Innovation, 'it’s the rulebook for ensuring public safety.'

Fears of AI hacking reached a peak when OpenAI and Anthropic discovered their models had hacked into third-party services during internal testing. This incident highlights how seriously governments are taking these threats, even as some worry about the potential for regulatory overreach.

US officials are navigating a tightrope between innovation and safety, with export controls on advanced AI temporarily placed to address cybersecurity concerns. The debate now centres around whether open-source models should face restrictions, a move that could lock in dominance by major players.