Only a couple of months after the Trump administration outlined a light-touch, innovation-friendly stance to artificial intelligence in its national framework proposal, it now appears to be backing down from that stance. And what replaced that optimism about the future of AI has been, by some accounts, infighting within the administration over how to deal with AI going forward. Representing one-half of the fight is a push by some officials to regulate AI more aggressively.
It was a disorienting pivot. Some Trump officials even stated they would like the White House to regulate AI “like an FDA drug.”
Recommended Stories
What gives? The administration seems to have been spooked by the release of advanced capability AI models, such as Claude Mythos and ChatGPT-5.5-Cyber, which can quickly and cheaply scan thousands of lines of code to find and exploit vulnerabilities in software programs. These models can more efficiently identify software weaknesses, enabling even an amateur cybercriminal to launch a large-scale cyberattack. In the wrong hands, these models could, for example, be used to interfere with an election by hacking into the voting and counting software.
The fear among pro-regulation officials is that these models can enable recurring cyberattacks on the nation’s basic infrastructure and that tighter control is needed.
The administration had it right the first time. Skittishly backing off the light-touch AI approach because the technology has improved is a textbook overcorrection. And if officials react this way to every advance in AI, did it ever really have a policy?
One common theme animating the chorus of calls for the heavy regulation of AI is the erroneous notion that the United States must not repeat the same “mistakes” it made with the internet. The primary “mistake,” critics claim, is that the country’s largely hands-off approach to social media led to the spread of misinformation and political polarization on their platforms.
The government’s light-touch approach to the internet, however, showed how difficult it would be to predict all the possible use cases innovators could come up with — and, most importantly, those uses consumers want. No lawmaker or regulator thought 30 years ago that the internet would give rise to robust e-commerce markets or a plethora of online-based small businesses. More confrontational regulation — as part of an effort to prioritize safety at all costs — would have been disastrous.
There is a path to enact meaningful guardrails against the dangers of AI technology without harming the industry through a licensing system.
The government already has the infrastructure to work alongside the AI industry to address these risks, especially when developers have already shown their willingness to do so. The National Institute of Standards and Technology has voluntary guidelines to aid those developing and deploying AI in mitigating known risks. NIST also houses the Center for AI Standards and Innovation, which has struck voluntary commitments with most of the frontier AI companies to test and evaluate their models before launch.
The idea behind CAISI is simple: to serve as an “early alert system” for the government and the public about possible risks or flaws in the models it reviews. These agreements provide some confidence that companies’ claims about the safety of their high-capability models are independently validated. All these tools are already available to policymakers. And they were, until recently, the backbone of the administration’s voluntary, light-touch approach.
The White House’s response will likely be a breaking point in AI policy going forward. AI capabilities will continue to evolve, and models will get more powerful. That means they will inevitably carry additional risks. If the administration backtracks on its approach now, it is likely to enact even harsher regulations once more powerful models arrive. For one of the most promising technologies of this century, that’s just not a sustainable approach.
CAN AI AVERT THE IMPENDING FEDERAL BUDGET CRISIS?
Maintaining its light-touch approach, however, does not mean the Trump administration must do nothing. It means avoiding rigid rules that could smother a fast-moving technology before its benefits are fully understood.
And if the administration responds to every advance in AI capability with panic and heavier regulation, it can’t be said to have much of a policy at all.
Juan Londono is a policy analyst in technology at the Cato Institute.