White House AI Model Vetting

The White House just compared AI regulation to FDA drug approval. Let that sink in for a second.

NEC Director Kevin Hassett confirmed this week that the administration is drafting an executive order to require pre-release safety reviews for new AI models — the same basic process that keeps untested drugs off pharmacy shelves.

The trigger was Anthropic's Mythos model, a system so capable of identifying network vulnerabilities that Anthropic itself flagged it as a potential global cybersecurity risk. The Commerce Department has already moved, expanding its voluntary testing program to include Google, Microsoft, and xAI alongside OpenAI and Anthropic. A full executive order is expected within the next two weeks.

I get the instinct. When an AI company has to warn the world about its own product, regulators are going to respond.

But here's what most people are missing in this debate. Drug approvals average 10 to 15 years. AI model cycles run in months. If pre-release vetting adds even 6 to 12 months to a deployment cycle, it doesn't just slow the race — it fundamentally changes who can compete in it.

Startups and open-source projects cannot absorb that compliance overhead the way a hyperscaler can. The unintended consequence of FDA-style AI regulation could be a massive consolidation of power at the very top of the stack.

The real question isn't whether AI should be vetted. It's whether the vetting framework will be designed to protect the public or protect incumbents.

Does pre-release vetting make AI genuinely safer — or does it mostly advantage the players who already have Washington offices?