Rules move slow. Models don’t.
That gap is the whole story of American AI policy right now, and if you’ve been trying to follow it from the outside, you’re not confused because you’re missing something technical. You’re confused because the situation is genuinely a mess.
Let me walk you through what’s actually happening, in plain language, because I think this matters more for regular people than most coverage suggests.
The Patchwork Problem
Here’s the setup. According to reporting from tech-insider.org, twenty-nine states have moved on AI legislation. What the United States does not have is a federal law covering AI.
Think about what that means practically. A company building an AI agent — the kind of software that books your appointments, screens job applications, or answers customer questions — is operating under a different set of expectations depending on which state the user happens to be sitting in. Not slightly different. Potentially contradictory.
The legal firm Spencer Fane framed this as models moving faster than the rules, which is an accurate description but also a bit gentle. It’s not just that the rules are late. It’s that when the federal level stays quiet, states fill the vacuum, and they fill it in twenty-nine different shapes.
Why this hits AI agents especially hard
Most AI regulation conversations imagine a chatbot sitting in a box, answering questions. Agents are different. An agent takes actions. It clicks things, sends things, decides things on your behalf.
That distinction matters enormously for policy, because “who is responsible” becomes a much harder question when software is doing rather than just saying. If an agent makes a bad hiring call or misfiles something with real consequences, the answer to who’s accountable might depend on state lines that the software has no concept of.
This is one of those cases where the technology and the legal framework are speaking different languages entirely.
Politics Enters the Room
ChinaTalk has written about how AI becomes a political crisis, and I think that framing deserves attention. AI policy isn’t staying in the technical lane. It’s becoming a political question, which means it will be shaped by the same forces that shape every other political question — timing, coalitions, elections, public mood.
For non-technical readers, this is actually useful information. It tells you that the eventual rules will not be written purely by people who understand transformer architectures. They’ll be written in a political process. That’s not a criticism. It’s just how democracies work. But it means predicting outcomes based on technical merit alone will mislead you.
Two Bigger Debates Running Underneath
The Carnegie Endowment for International Peace has published on two threads that connect to all of this.
- Compute and international coordination. Their work on a “Compute Coalition” looks at building the future of AI in the free world — meaning the question isn’t only domestic. Who has access to the hardware that trains these systems is a geopolitical matter, and any domestic rulemaking sits inside that larger frame.
- Labor. Carnegie has also laid out three different views on the future of work under AI. Three views, not one consensus. That’s a signal worth reading carefully — the people studying this professionally don’t agree on what happens to jobs.
I bring up that second point because it’s where most people actually feel this issue. Not in compliance frameworks. In whether their work changes. And the honest answer from serious researchers is that there’s real disagreement, not a settled forecast being hidden from you.
What I’d Actually Tell a Friend
If someone asked me how to think about this without a law degree or a machine learning background, I’d say three things.
First, expect inconsistency for a while. A patchwork of state rules with no federal layer above it does not resolve quickly. If you’re using or building AI agents, the compliance picture in 2026 looks fragmented.
Second, notice who’s shaping the conversation. Law firms, think tanks, state legislatures, and international policy groups are all pulling at this from different angles with different incentives. Reading the source tells you a lot about the argument.
Third, don’t mistake disagreement for ignorance. When researchers offer three views on AI and labor, that’s intellectual honesty about genuine uncertainty. The people telling you they know exactly how this plays out are the ones I’d trust least.
Where That Leaves Us
The core tension is simple to state and hard to fix. Software iterates in weeks. Legislation iterates in years. Twenty-nine states have decided not to wait, and the result is a national system with no national rule.
Whether that’s a temporary gap or the new normal is the question everyone in this space is quietly betting on. My advice? Watch the states. That’s where the actual rules are being written right now, whether or not Washington catches up.
🕒 Published: