US Government Is Weighing Regulating Public Access to AI Models

So are the states. Why? The latest models are very adept at identifying security issues in existing code and exploiting them. The panic began with Anthropic's Claude Mythos model in early April. There are a lot of reasons software is vulnerable, including the fact that probably too much of it has been hanging around for up to 40 years. We aren't great at cleaning up unused code (think software updates on your computer, or the stuff at the bottom of your closet). We tend to build on existing code, patch it, extend it, convert it, etc. Companies are having the same issue with AI agents today. This problem is sometimes referred to as technical debt.

Some equate the risk to nuclear weapons in an attempt to get everyone's attention. On one hand, that feels disrespectful to the victims in Japan and conjures up horrific images. On the other, it speaks to the magnitude of power a nation or entity would hold if it possessed the technology to disrupt operating systems in our energy infrastructure, banks, governments, and more. I'm just finishing Louise Penny's latest novel, The Last Mandarin. Pick up a copy if your imagination is failing you right now.

My take: There are reasons to be concerned. The capabilities of foundational models are advancing rapidly, and we don't have the government infrastructure to address AI's impact quickly enough. Executive orders can be a stopgap, but we need a thoughtful approach that balances the needs of a broad set of stakeholders — the interests of government, private entities, and citizens are not fully aligned. A handful of entrepreneurs and investors are poised to generate unprecedented wealth. Trust is also absent: citizens have little faith that the US government or tech companies will do the right thing. According to a Pew Research study from September 2025, just 17% of Americans trust the federal government to do what is right most of the time — a seven-decade low. The Edelman Trust Barometer reports that US citizens are nearly 3x more likely to reject AI than to accept it.