Pentagon's Tech Chief Says AI Labs That Don't Cooperate Can Face Export Controls, and Anthropic Already Did

The man who now runs Washington's relationship with the big AI labs has said the quiet part out loud. Emil Michael, under secretary of war and the Pentagon's chief technology officer, says the US government can hit AI companies that don't cooperate with export controls. He told The Information it has already happened once.

That matters because export controls are one of the heaviest levers Washington has over a company that sells around the world. They decide who can buy your product and where it can go. And Michael is no longer just a Pentagon official. He spoke the day after the first full meeting of the White House's new AI task force, the Super Intelligence Force, where as a vice chair he is responsible for the government's dealings with the labs and for how they report incidents.

Two hours

A White House official had already floated export controls as one possible stick. Asked whether they were an appropriate way to keep companies in line, Michael said yes. He pointed to past limits on exporting encryption and on foreign nationals using certain technologies, and said the executive branch has plenty of other tools, including enforcement by the Justice Department.

Then he offered an example. According to Michael, Anthropic refused to pull back a model after the government learned it had been jailbroken, and within two hours the company was under export controls. "I think you can count on this administration, this president to act with urgency should there ever be a real issue," he said. That is his account; he did not say which model was involved.

He was less patient with AI companies that ask to be regulated, the argument that the government should stop them because they could do something dangerous, so rivals don't get there first. Michael said it made little sense and summed up the pitch as: "I'm the drunk driver in the car. You've got to pull me over." The jab lands squarely on Anthropic, which recently drafted its own federal AI safety bill. In his view, existing laws are "more than sufficient," based on his own reading and on what FTC chair Andrew Ferguson has told him.

Still shut out

Anthropic remains out of the Department of War, as the department said six months ago it would be, after the Pentagon labelled it a supply chain risk. An appeals panel let that blacklisting stand, and Anthropic has now asked the full appeals court to rehear the case. Michael expects the government to win.

He still talks to the company, and saw CEO Dario Amodei at a White House event a few weeks ago. On the task force, he says, Anthropic will be treated fairly and even-handedly like the other labs.

Working with it inside the department is another matter. Michael objects to the idea that a model has feelings and can overrule a company's founders, which he says Anthropic builds into its software through the constitution attached to its models. In his words: "If you don't want humans to lose control, you can't give models the control over humans." In the military, a model cannot come between a commander and their superior. Unless Anthropic changes that, he said, there is nothing to discuss. Elon Musk's company and Google signed the same terms of use, covering all lawful uses.

AI inside the Pentagon

Inside the department, AI use has taken off anyway. About 1.7 million of its 3.5 million employees now use AI, Michael said, up from 80,000 a year ago. In June the Pentagon's AI platform had 1.5 million users.

He favours American open-source models, naming Nvidia's Nemotron, Reflection AI and Thinking Machines, though he concedes none is yet at the frontier. He worries a little about Chinese open-source models becoming dominant in the US, because their makers answer to the Chinese government.

The task force's to-do list is concrete: an incident response framework, plans to harden critical infrastructure, and a review of the laws that already apply to AI companies. Whether the country needs new ones is a separate question, handed to the President's science advisers.

Some offers on this page may be paid placements or contain affiliate links.