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

This week, the tech world is abuzz with the revelation that the White House has finalized new AI rules, but with an unusual caveat: they won't be publicly released. This decision has sparked considerable debate and concern among industry experts and the public alike. Kevin Roose, a tech columnist at the New York Times, and Casey Newton from Platformer, delve into the intricacies of these regulations, their potential impact on AI development, and the broader implications for safety and innovation. They also sit down with Chris Painter, President of METR, a non-profit focused on AI research and testing, to discuss the growing concerns around AI model misalignment and the challenges of ensuring responsible AI deployment.

The White House's Covert AI Framework

Roose and Newton kick off the discussion by highlighting the unprecedented secrecy surrounding the White House's new AI framework. While the administration has finalized its guidelines for testing "new frontier AI models" from major American AI companies, the details remain under wraps. Newton points out the irony of this approach in a democracy, where transparency is typically paramount for new regulations. The framework reportedly grants the government a 30-day window to access frontier models before their public release, allowing for testing and evaluation in "high-security environments." However, the voluntary nature of this submission process, as described by the Trump administration, raises immediate questions about compliance and enforcement. Newton speculates that companies might resort to creating parallel models—one for testing and another for continued use—to circumvent potential delays. Roose further emphasizes the dynamic nature of AI development, where models are constantly updated, making a 30-day freeze for testing impractical and potentially counterproductive.