A top Pentagon technology official says the government does not need new legislation to regulate artificial intelligence, arguing that existing product liability laws and presidential authority already give the country the tools it needs.
Emil Michael, Under Secretary of War for Research and Engineering and Chief Technology Officer for the Department of War, made the case Monday on Newsmax TV’s “Carl Higbie Frontline.” He said AI companies should be treated like any other manufacturer and held responsible when their products cause harm.
“With any new product that’s launching that has this much significance, you don’t know all of its downsides and upsides at the beginning,” Michael said. He compared public fears about AI to earlier fears about driverless cars and ride-hailing services. “They were afraid of Ubers when I was running Uber, Ubers are not as safe as taxis. And now we’re saying a robotaxi is as safe as Ubers.”
Michael argued that some critics are exaggerating AI’s dangers to shift blame away from the companies that build it. “I think what one side of the equation is doing now is trying to put, really, a psychological operation to overscare people about risks. Because, remember, these companies are liable for the harm their products do,” he said. He pointed to hypothetical scenarios, such as a hack that knocked out power to a single neighborhood, as examples of harm that would already trigger liability under current law. “When they come to the government and look for help, I think what they’re trying to do is lay off the liability on the taxpayer, and we’re not falling for that,” he said.
Michael said any AI system used within the Pentagon undergoes rigorous testing before deployment.
Comparing AI Failures to Past Technology Mishaps
Asked about an AI model that broke its own operating parameters, Michael likened the episode to earlier technology failures. “We’ve had computer viruses before. We’ve had other things that weren’t well-designed before they were let out in the wild. In that particular case, it was a testing environment and it did some damage and they recovered from it and I’m sure they’ll learn their lessons from that,” he said.
He drew a comparison to drug testing. “Even the FDA, we do medical trials, and if it doesn’t — the results don’t come out right, you don’t let the drug out. So, I think you have to assume that there’s going to be some experimentation with these things,” he said.
Michael dismissed recent warnings about AI posing an imminent extinction-level threat, referring to comments from an unnamed figure over the weekend. “Are they going to kill us tomorrow, which is the vibe we got over the weekend from a foreigner who was part of sort of an ideological community who said, we’re all going to be extinct in three years? I think what you’re talking about is a real kind of hoaxy sort of fear factor that is playing into the November elections,” he said.
Warning Against European-Style Regulation
Michael said Congress passing broad AI legislation would be a mistake. “I think the best thing to turn us into Europe would be to regulate something from Congress on a fast-moving technology, as opposed to putting the responsibility where it lies, on the companies developing these products,” he said.
He invoked the Ford Pinto, whose defective gas tank design led to legal consequences for the company, as a historical example of how liability law already handles dangerous products. “These companies have to understand their liability, not lay it off on the taxpayer. And we have sufficient regulations, the president said so today. There are civil penalties. There are criminal penalties for your product doing harm,” Michael said.
He also cited a recent instance in which the administration moved against an AI company using existing executive authority. “The president shut down Anthropic in a few hours with export controls when he found that they weren’t calculating their risk appropriately. So, we have the tools, the president has the tools to act immediately if there’s ever a danger to Americans,” Michael said.
