The government wants to rein in powerful AI, but there are downsides

The government is stepping up its push to rein in powerful AI models that it believes could endanger national security. On June 26, Representatives Josh Gottheimer, a New Jersey Democrat, and John Moolenaar, a Michigan Republican, introduced the bipartisan Cloud Security Act, which would have AI companies flag suspected “misuse” of their platforms to the government. The legislation, advocates say, would close a critical loophole in current export rules designed to prevent rival countries like China from acquiring advanced AI chips. (Fast Company has reached out to Gottheimer and Moolenaar for comment.) While there are already sharp limits in place on the sale of these chips, lawmakers want to empower tech companies to more freely report to the federal government suspicions that a foreign actor is using their products to create advanced AI models. The bipartisan bill comes as the Trump administration is moving forward with plans to limit public access to leading American AI models. In recent days, federal officials worked with OpenAI on a deal that would sharply restrict which customers are granted access to its upcoming ChatGPT 5.6 model. That move came just weeks after the Commerce Department pushed Anthropic to roll back its release of Fable 5, a version of its powerful Mythos model, citing security concerns. The move effectively prevents Anthropic from selling its most advanced product. The company has observed the order and pulled the model offline. There are real reasons, of course, to worry about the implications of making ever-more dextrous AI models publicly available, and the very real possibility that they will be used by bad actors. Still, there are also reasons to be concerned about the growing movement by the federal government to limit access to certain AI models. Even if these limitations do help protect national security in some ways, a blunt and sudden curtailment also runs the risk of leaving other organizations without technology that could help them protect themselves. In effect, these limits also allow the government to pick and set the terms by which America’s leading AI companies can access the consumer market. “Concern number one is if we’re going to take the defensive capability out of the hands of U.S. government, of local government, and of U.S. industry, and we’re going to tie our hands, at least one hand, behind our back, we better have a really good and clear justification for that because the downside risk is enormous,” California Democratic Representative Sam Liccardo tells Fast Company. “If these are really advanced models that are going to help us defend ourselves against the very advanced models to come, we’re nearly out of time.” Liccardo was one of four lawmakers who on June 18 sent a letter to the Commerce Department about the sudden new limits on Anthropic’s Fable 5, including questions about whether the agency bypassed normal regulatory processes to take the model offline and, in effect, targeted Anthropic. Responses to those questions, and more, were due at the close of business day on June 26 but never came, Liccardo’s office tells Fast Company. Much ado has been made of the Trump administration’s ongoing battle with Anthropic, but Liccardo says the concern is much larger. If adversaries are building increasingly powerful models, he fears that many American institutions won’t have access to powerful U.S. models that they may need to defend themselves. Even more worrisome is that standards for how to enact those systems aren’t clear, even to lawmakers. “If we have an export licensing regime without any review or substantial oversight until months or even years later,” Liccardo warns, “that oversight becomes worthless and meaningless in an industry where progress is measured by days and weeks.”