The White House’s Secret A.I. Rules + The State of Model Alignment With METR’s Chris Painter + The Final Hot Mess Express

In one of the strangest developments in recent AI regulation, the White House has finalized its framework for testing new frontier AI models — but has not released it publicly. The rules were briefed privately to representatives from OpenAI, Anthropic, Google, and other companies, but the rest of the world has been left to piece together what they contain from leaks and reporting. One AI lab insider compared the situation to "regulatory Calvinball," where the rules are made up as you go along.

The Secret AI Testing Framework

According to reporting from Axios, the framework gives the government a 30-day window to access frontier models before they are released publicly. Companies developing closed-source frontier models with advanced and potentially dangerous capabilities would submit them to the government, which would then run evaluations in "high security environments." Multiple administration offices would be involved rather than a single agency. The Trump administration has stated that participation is voluntary — though as the hosts note, this is "voluntary in the same way that paying your taxes is voluntary."