The DOJ Task Force Built to Sue States Over AI Just Picked Its First Target. It Wasn't a Safety Law.
On April 24, the Department of Justice filed its first substantive intervention in a lawsuit challenging a state AI law, backing xAI's case against the Colorado AI Act. The move follows through on Executive Order 14365, which President Trump signed on December 11, 2025, directing DOJ to stand up a litigation task force to identify and challenge state AI statutes that conflict with federal policy, and directing agencies to use funding leverage against states that keep laws Washington deems unduly burdensome. Four months after that order, the task force has its first real case on the board.
A civil-rights law, framed as a civil-rights problem
What makes the Colorado intervention notable is the target. The Colorado AI Act is not a catastrophic-risk disclosure law in the mold of California's SB 53 or Illinois's audit mandate — it's an anti-discrimination statute, requiring companies to test AI systems used in hiring, lending, housing and similar high-stakes decisions for discriminatory outcomes. DOJ's brief argues the opposite of what the law's sponsors intended: that its anti-discrimination provisions actually force developers to weigh race, sex and religion when evaluating model outputs, turning a civil-rights law into evidence of the civil-rights violation DOJ says it exists to prevent. That framing echoes the executive order's language almost exactly, and it is a different legal theory than the one most observers expected the task force to reach for first — a straightforward federal-preemption argument against a safety-disclosure regime, not a civil-rights argument against an anti-bias one.
The safety-framework bills it didn't touch, yet
The task force's choice of target leaves the other family of state AI laws untouched for now, even as it keeps growing. Pennsylvania's House Communications and Technology Committee heard initial testimony on October 19 on two bills, HB 2800 and HB 2829, that would require large AI developers to publish safety frameworks and submit to independent audits — core architecture borrowed directly from California's SB 53, Illinois's audit law and New York's RAISE Act, the same family of statutes Massachusetts is simultaneously negotiating its own version of in a conference committee fight splitting Anthropic from OpenAI and Google. None of those laws has drawn a DOJ court filing yet. But the Colorado intervention establishes that the administration will use federal court, not just the bully pulpit, to contest state AI statutes it considers inconsistent with its deregulatory posture — a precedent the safety-framework states now have to litigate around even though Colorado's law is substantively different from theirs.
That distinction matters for how durable any given state law is likely to be. A civil-rights argument against an anti-discrimination statute doesn't automatically transfer to a disclosure-and-audit statute like Pennsylvania's or Massachusetts's — DOJ would need a different legal theory to challenge those, most likely a direct preemption claim once a safety-framework bill actually becomes law somewhere and a developer wants to contest it. Pennsylvania's bills are nowhere near that stage: Monday's hearing was an initial airing of testimony, not a committee vote, in a chamber already juggling more than a dozen other AI-related proposals this session.
What the Colorado case does establish, regardless of its narrower subject matter, is that the litigation task force EO 14365 created is no longer just a mandate on paper. States writing the next generation of AI safety bills — and Pennsylvania's sponsors, who have argued plainly that "currently, AI has no federal regulatory standards," are explicitly building toward one — are now doing so knowing the federal government has shown it will go to court over state AI law it doesn't like, even if the first case it picked wasn't the kind of law most people expected it to pick.