Full Breakdown
Trump Administration Considers Voluntary AI Model Vetting Order
5/21/2026, 4:21:17 AM
Policy Shift from Deregulation to Oversight
President Donald Trump’s early-term AI policy emphasized deregulation, including repeal of President Biden’s AI safety order and the “One, Big, Beautiful Bill” Act that blocked state AI laws. By late 2025 he threatened states that kept their own rules, and in 2026 introduced a vague federal AI framework. Leaked drafts now propose a voluntary framework for vetting frontier AI models.
Key Actors
The draft cites President Donald Trump as initiator; Vice President J.D. Vance renounced regulation at a 2025 AI summit. The Center for AI Standards and Innovation (CAISI) is mentioned as a possible overseer. Anthropic CEO Dario Amodei, and tech firms Google DeepMind, xAI, Microsoft, and OpenAI have engaged with the administration on testing.
Timeline
- Early 2025: AI Action Plan calls for reduced regulation.
- Late 2025: Executive order threatens states; vague national framework announced.
- April 2026: Anthropic releases Claude Mythos, exposing thousands of software vulnerabilities.
- May 2026: Draft order splits into “cybersecurity” and “covered frontier models” sections, offering a 90-day voluntary check-in.
- June 2026: Polls show strong public backing for pre-release testing.
Public Opinion
Morning Consult reports 47 % of Republican voters “strongly support” the plan, 5 % “strongly oppose,” and over 70 % favor mandatory independent testing. An Institute for Family Studies poll finds >80 % of Americans support the plan; Gallup 2025 data show 80 % of U.S. adults prioritize AI safety over rapid innovation.
National Security Impact
Claude Mythos showed advanced coding and reasoning, revealing tens of thousands of previously undetected software bugs. The Pentagon has blacklisted Anthropic for refusing unconditional access, prompting officials to view the model as a catalyst for tighter AI oversight.
Official Statements & Responses
Leaked drafts outline a “cybersecurity” section that hardens federal cyber defenses without directly regulating AI, and a “covered frontier models” section that creates a voluntary framework. Developers receive a 90-day window to submit models for government vetting. No formal announcement or working-group creation has been made.
Criticism & Opposition
Analysts argue the voluntary framework lacks enforcement power, reducing company incentives to comply. Critics warn the 90-day check-in may be insufficient for rapid deployment of high-risk models.
Conflicting Reports & Gaps
Sources disagree whether the framework will stay voluntary or become mandatory; the draft uses “voluntary” while polls demand mandatory testing. The overseeing agency—potentially CAISI—is not definitively identified, and the final order language remains unreleased.
What’s Next
The administration is expected to finalize the order and may form a working group of government officials and tech leaders to formalize vetting procedures. The 2026 midterm elections and ongoing public pressure could shape the scope and enforceability of AI oversight measures.
