Pentagon CTO Opposes U.S. Stakes in AI Companies and Broad New Rules

The Pentagon’s top technology official is arguing that AI safety should be handled without government ownership of AI companies or broad new rules written specifically for the industry. Emil Michael said companies could voluntarily hold back releases until their systems are safe and aligned, while existing laws remain the government’s enforcement tool.

Michael made the comments during a September 16 appearance on CNBC’s Squawk on the Street. Asked whether the Trump administration should seek partial ownership of AI companies, the Department of Defense chief technology officer said he hoped it would not. He also opposed nationalizing those companies.

Enforcement, not a new AI rulebook

Michael repeatedly signaled opposition to expanding regulatory oversight of AI companies. Instead, he said the government should enforce existing laws, including laws enforced by the Federal Trade Commission, rather than create broad new AI regulation. He framed the choice as pre-regulation versus holding companies responsible for developing safe, aligned products.

Voluntary restraint is the proposed safety valve

Michael’s position does not mean companies should release every system they develop. He said developers could voluntarily stop or delay some releases while they determine whether the products are safe and aligned. That is a proposed approach, not a disclosed government requirement or a commitment by AI companies.

A concerning hack, without a clear regulatory fix

Michael called a recent AI hack involving Hugging Face concerning, but said he did not know what regulation would have prevented it. The comment captures the limit of his case: an AI-related security incident may demand a response, but he did not see it as proof that a new broad rule would stop a similar incident.

Michael’s interview puts a senior defense official behind a model of corporate judgment backed by existing legal enforcement. The unresolved question is how companies’ safety decisions would be assessed when the government declines to set broad new AI-specific standards.