Trump is about to sign an AI executive order as early as tomorrow (Thursday) that would ask frontier AI labs to share their models with the government 90 days before public release.

But here's the catch: it's voluntary.

The cybersecurity section is mandatory. Pentagon, hospitals, banks, and critical infrastructure will be required to harden their systems against AI-powered attacks. That part has teeth. The frontier model section is what they're calling a 'voluntary framework.'

This represents a massive shift from Trump's original 'no regulation' AI stance. Remember JD Vance's speech in France where he basically said the market would figure it out? That vision is dead.

There's been internal White House infighting over the approach. Some advisors wanted mandatory access to all frontier models. Others wanted the government to stay completely hands-off. The compromise is voluntary model sharing with CAISI (the AI safety institute) doing the oversight.

The order also mandates wider AI deployment across government systems within 30 days. Federal agencies will be required to identify where AI can replace manual processes. Translation: a lot of government workers are about to discover their jobs can be automated.

This directly affects Anthropic (Claude Mythos is specifically named), OpenAI, Google, and any lab working on frontier capabilities. It's the first time this administration has acknowledged that frontier models might need ANY government review.

The voluntary nature is probably intentional. Trump gets to say he's addressing AI safety without actually restricting innovation. Labs get to participate or not. If something goes wrong, he can point to the voluntary framework and say companies should have engaged.

The real test will be whether frontier labs actually participate. Anthropic and OpenAI will probably say yes. Google might hedge. The whole thing feels like security theater designed to look like action without actually constraining anybody.