Colorado was supposed to be the proof that America could regulate AI the serious way, the way Europe did.
It isn’t anymore. The state just gutted its own landmark AI law under direct federal pressure, and almost nobody outside the legal trade press has noticed.
Here’s the date everyone’s still citing, including most of the compliance guides circulating right now: June 30, 2026, the deadline for Colorado’s original AI Act to take full effect. That date is dead. It was real once. The law, passed in 2024, got delayed once already, from February to June of this year, after Governor Polis and the legislature couldn’t agree on amendments. But the delay was supposed to be a pause, not an ending. Then a federal court stepped in. xAI, Elon Musk’s AI company, sued to block the law, arguing it forced AI developers to embed the state’s preferred views into their products. The Department of Justice didn’t sit on the sidelines. It intervened on xAI’s side, through an AI Litigation Task Force built specifically to challenge state AI laws on constitutional grounds. In April, a federal judge stayed enforcement of Colorado’s law entirely.
Three weeks later, the Colorado legislature didn’t fight. It rewrote the whole thing. The new law, signed in mid-May, strips out nearly everything that made the original serious: gone is the duty of care developers owed to prevent algorithmic discrimination, gone are the mandatory risk management programs, gone are the impact assessments that would have forced companies to document how their systems might harm people in hiring, lending, healthcare, and housing decisions. What’s left is a much thinner disclosure-and-rights framework. Tell people when they’re talking to an AI. Give them a way to challenge a bad automated decision after the fact. That’s not nothing, but it isn’t what Colorado built in 2024, and it isn’t what the rest of the world was watching to see if America could do.
That matters because Colorado wasn’t just any state. It was the one state that had actually tried to import the European model: a real, codified duty of care, the same conceptual architecture sitting underneath the EU AI Act. Other states have passed AI laws, but Colorado’s was the only one with real teeth, the only one treating algorithmic discrimination as a thing companies had an affirmative obligation to prevent rather than just disclose. When the one state running that experiment folds it under direct legal pressure from the federal government, that’s not a local story. That’s the answer to a question this whole industry has been asking since the EU AI Act passed: will America build its own version of that, state by state, since Congress won’t? Colorado just told you no.
Watch what replaces the watching. The new law takes effect January 1, 2027, and unlike its predecessor, this one is expected to actually hold, because the legislature won’t reconvene until the deadline has already passed. There’s no more room left to delay it again, because there’s nothing left in it worth fighting over. That’s not stability. That’s a law getting simple enough that nobody needs to keep arguing about it.
Here’s the part worth sitting with longer than the headline. This didn’t happen because voters changed their minds about AI risk, or because new evidence emerged that algorithmic discrimination wasn’t a real problem. It happened because one company sued, the federal government backed that company’s lawsuit, and a state legislature concluded it couldn’t win that fight and rewrote its own law to make the fight go away. Whatever you think about whether Colorado’s original law was workable, the mechanism that ended it should give every state legislature pause: when a federal task force built to challenge state AI laws picks your law as the target, you do not get to decide this on the merits anymore. You get to decide how much of your own law you’re willing to keep before the lawsuit decides it for you.
The EU passed a real, binding, risk-based AI law and is holding to it. Colorado tried to build the American version of that and just watched it collapse under its own federal government’s pressure, three weeks into what should have been its first real test. If that’s the pattern other states read out of this, the era of state-level AI governance with actual teeth may have ended before most people noticed it started.
“The Faust Baseline Codex 3.5”
micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
”AI Baseline Governance”
Purchasing Page – Intelligent People Assume Nothing
Unauthorized commercial use prohibited. © 2026 The Faust Baseline LLC
