AI #167: The Prior Restraint Era Begins
White House now demands veto power over AI model release decisions.
The era of training and releasing frontier AI models at will appears to be over. According to Zvi's analysis on LessWrong, the White House now requires advance review of release decisions and has already exercised a veto to block expansion of access to the Mythos model. Hassett's comparison to FDA-style regulation signals a shift toward slow, centralized approval processes that could hamstring US AI innovation without parallel constraints on China. The administration is also engaging in talks with China to coordinate model access restrictions, which may offer a more balanced path forward.
In parallel, Anthropic continues its explosive growth, leasing SpaceX's Colossus 1 supercomputer to immediately expand usage limits. Elon Musk has begun speaking positively about Anthropic's motivations, even as testimony in the Musk v. OpenAI trial rehashes old details. Anthropic's ARR has reached $44 billion, with potential fundraising at a $900 billion valuation. Other developments include GPT-5.5 Instant, faster Gemma 4, a ProgramBench where all models score 0%, and Meta facing a new copyright lawsuit. The week also saw new AI regulations in Maryland and Connecticut, a union vote by DeepMind workers, and warnings about AI risk from Jack Clark.
- White House vetoed expansion of Mythos model access, setting precedent for prior restraint on frontier AI.
- Hassett compared regulation to FDA approval, threatening to slow US AI development unless China follows suit.
- Anthropic leased SpaceX's Colossus 1, hit $44B ARR, and may raise at >$900B valuation.
Why It Matters
Professionals face a new regulatory paradigm that could delay model releases and reshape AI competition dynamics.