The White House is tightening control over AI testing by ordering CAISI, the Commerce Department unit, to halt public reports on frontier AI model reviews as it implements an executive order focused on stronger security considerations in model evaluation.

This pause on public disclosures follows guidance from National Cyber Director Sean Cairncross and aligns with the administration’s push to move evaluations from open publishing to internal, national-security–driven channels.

The order accompanies a framework allowing voluntary government access to frontier models before wider release, with a 30-day government review period during which the Treasury, NSA, and CISA oversee assessments.

Investors face increased regulatory ambiguity as reputational gains from transparent participation may erode if results stay internal, prompting some firms to reassess CAISI cooperation and disclosures.

Anthropic’s Mythos had a staged, limited release culminating in a broader public version with added guardrails released recently, illustrating the tension between openness and safeguards.

CAISI historically runs voluntary testing with major developers like OpenAI, Anthropic, Google DeepMind, Microsoft, and xAI, without mandatory licensing or pre-approval requirements.

OpenAI and Anthropic have long-standing ties to CAISI, dating back to its origins as the AI Safety Institute, with calls to strengthen CAISI’s mandate.

As of May 2026 CAISI evaluated over 40 models and partnered with Google DeepMind, Microsoft, and xAI, a collaborative, open-science approach now facing disruption under the new framework.

Implications include higher regulatory frictions for U.S.-based AI firms, potential competitive disadvantages to international players, and a shift for CAISI staff toward a more national-security–driven role.

The order reflects a blend of support from Treasury Secretary Scott Bessent and the National Cyber Director with questions about CAISI’s role and public transparency.

The executive order is framed in the context of Anthropic’s Mythos capabilities and broader debates over AI safety, transparency, and national security.

The shift signals a move from open science toward a national-security asset model, reducing independent verification for investors, researchers, and the broader industry.