White House SIF Co-Chair: "We Will Control AI Companies That Do Not Cooperate with Government"

Emil Michael, a U.S. Department of Defense official and co-chair of the White House Superintelligence Force (SIF), said the government would use different administrative measures, including export controls, to control AI companies that refuse to cooperate with the government.

Michael was the official who pushed to label Anthropic a "supply chain risk company" after the company refused to remove safety rules from its AI system.

The technology news outlet The Information reported on October 19, local time, that Michael made these comments in an interview with the publication.

Michael said the Trump administration has a duty to control AI companies such as Anthropic.

So far, the Trump administration has mainly relied on voluntary steps to oversee AI. For example, major AI companies signed a non-binding agreement after a White House meeting late last month. This means the agreement does not legally force companies to follow its terms.

However, Michael said the government should use stronger checks and controls on AI companies when needed.

He pointed to export controls introduced last June that stopped people outside the United States from using Anthropic's AI model as an example.

"We have many tools available to us. These range from limiting the export of technology that could help our enemies, to strict export controls, to stopping foreign nationals from using the software," he said. "You can trust that if a real problem arises, the administration and the president will act quickly."

He also opposed treating AI models as if they were people. As an example, he pointed to Anthropic's policy that allows its Claude model to end conversations with users who intend to cause harm.

"If we don't want humans to lose control, we must not give models the power to control humans," he said.

He added that he would not allow Anthropic's technology to be used by the Department of Defense as long as "such moral principles are built into the software."

Michael criticized leading AI companies that call for stronger safety rules. He compared them to drunk drivers who want someone else to stop the car instead of stopping it themselves.

He also agreed that the national security risks linked to AI are growing.

"It is a bit concerning that China's open-source models are becoming dominant in the U.S., because those companies are controlled by the Chinese government," he said.

He said one way to respond to the growing use of these models in the United States would be to check their model weights, the internal values that help determine how AI systems work. However, he said he would rather address the issue by supporting the growth of American open-source alternatives, such as Reflection, Thinking Machines, and NVIDIA's NeMoTron.

However, he acknowledged that none of these companies has yet reached the most advanced level of AI development.

Michael also said that industrial spying is a concern.

He referred to a recent case in which an OpenAI employee who was a Saudi national installed a wiretapping device. "Not all foreign nationals are bad, but there is a certain amount of risk associated with foreign nationals," he said.

He argued that companies should be more alert to these risks rather than relying on the government to impose even more controls.

Please note: This article was translated by AI and may contain errors.