White House Blocks Anthropic's Mythos Expansion, Signals Shift to AI Prior Restraint Regime
Key Takeaways
- ▸White House ordered Anthropic to cancel expansion of Mythos access under Project Glasswing, the first concrete application of government model deployment approval authority
- ▸U.S. government is considering implementing a formal prior restraint regime requiring approval before any frontier AI model release
- ▸Policy reversal represents dramatic shift from previous hands-off frontier AI approach, allegedly driven by cybersecurity concerns and the existence of Mythos
Summary
The White House has ordered Anthropic not to expand access to its Mythos model as part of Project Glasswing, marking the first concrete application of what appears to be a new AI governance framework. The administration is now seriously considering implementing a formal prior restraint regime that would require any company releasing a highly capable frontier AI model to obtain government approval before deployment. This represents a dramatic reversal of the U.S. government's previous hands-off approach to frontier AI policy and indicates growing concerns about security risks, particularly regarding cyber attacks.
The blocked Glasswing expansion was intended to provide key companies—including European firms—access to Mythos for security purposes. Analysts note that while the immediate impact may be minimal, the precedent of informal White House veto over AI deployments is troubling, as it enables ad-hoc decision-making that could favor politically connected parties over evidence-based policy. The government is reportedly establishing an AI working group of tech executives and officials to develop procedures for reviewing frontier model releases.
Experts including Neil Chilson and Dean Ball warn that the emerging approach lacks formalized procedures and safeguards, creating an "ad-hoc" regulatory environment that could be less accountable than formal rules. The shift comes amid heightened concerns about potential "hackastrophe" scenarios where defenders need early access to advanced models to counter sophisticated cyber threats, providing the rationale for the policy reversal.
- Ad-hoc implementation without formalized procedures raises concerns about corruption, insider favoritism, and unpredictable governance
Editorial Opinion
While a well-designed prior restraint regime for frontier AI models could be justified by legitimate safety and security concerns, the ad-hoc nature of the emerging approach is genuinely problematic. The government's failure to prepare formalized procedures in advance—despite years of warnings—has resulted in reactive, informal decision-making that lacks transparency and accountability. This pattern of crisis-driven governance sets a precedent for arbitrary executive power over AI deployment that could be easily abused, regardless of current intentions.
