DOJ Just Made Colorado's AI Anti-Discrimination Law a Federal Constitutional Fight

On April 24, the Justice Department moved to intervene in xAI’s challenge to Colorado SB24-205, the state’s algorithmic discrimination statute. The filing argues the Colorado law violates the Fourteenth Amendment’s Equal Protection Clause by forcing AI companies to prevent unintentional disparate impact while carving out an exemption for algorithms designed to advance “diversity” or redress “historic discrimination.” That carveout, the DOJ says, is the constitutional defect. You cannot ban one form of disparate impact and bless another.

Assistant Attorney General Harmeet Dhillon called the law’s structure “woke DEI ideology” baked into a regulatory regime. Assistant Attorney General Brett Shumate framed it as a national security issue. Rhetoric aside, the legal theory is sharp and it has teeth. xAI filed the underlying suit on April 9. Two weeks later the federal government joined. That timeline matters.

What Colorado SB24-205 Actually Requires

The statute targets “developers” and “deployers” of AI systems used for consequential decisions. Mortgage lending. Student admissions. Job-candidate selection. Healthcare access. Insurance pricing. Anything that hits a person’s pocketbook, their housing, or their future.

Developers must complete impact assessments, document training data choices, and disclose known risks of algorithmic discrimination. Deployers must run their own assessments, notify affected individuals when AI participated in a decision, and offer an appeal mechanism. Both sides face civil penalties enforced by the Colorado Attorney General. The compliance lift is real, and Colorado built it with the Connecticut and California models in mind.

The carveout the DOJ is attacking sits in the definition of “algorithmic discrimination” itself. The statute exempts AI systems used to “expand an applicant pool to increase diversity” or to “address historic discrimination.” That language was drafted to protect affirmative-action-style algorithms from triggering disparate-impact liability. The DOJ’s read: you cannot prohibit discrimination in one direction while protecting it in the other. Equal Protection runs both ways.

Why Federal Intervention Changes the Math

A private company suing a state over an AI law is interesting. A private company plus the United States, with the Civil Rights Division and the Civil Division both signing the brief, is a different animal. Three things shift the moment DOJ enters:

Resource asymmetry. Colorado’s AG office is competent, but it is not built to fight a coordinated federal challenge with unlimited briefing capacity. The state will spend real money defending a law it passed in 2024.

Preemption pressure. Even without a formal preemption claim, federal participation signals to other states that their AI bias laws are now on the constitutional radar. New York, Illinois, Texas, and California all have similar frameworks in various stages. A loss in Colorado ripples.

Settlement room drops to zero. Colorado cannot quietly narrow the statute through enforcement guidance the way it might have if only xAI was at the table. DOJ wants a ruling, not a fix.

The Equal Protection Theory Is Stronger Than the Politics

Strip out the press release language about “radical far left worldviews” and the legal argument is straightforward. Under Students for Fair Admissions v. Harvard (2023), the Supreme Court held that race-conscious admissions programs violate Equal Protection even when they aim at remedying historical inequality. Colorado wrote a statute that effectively requires the same race-conscious algorithmic design that SFFA rejected, and exempted it from a disparate-impact prohibition that applies to everything else.

If a court takes that comparison seriously, the carveout falls. And if the carveout falls, Colorado is left with a statute that prohibits any AI system causing disparate impact on protected classes, which is itself a problem because Title VII, ECOA, and the Fair Housing Act already do that work at the federal level. The state law becomes either unconstitutional or redundant. Neither outcome is good for Colorado.

What This Means for Compliance Teams

If your company built or is building a Colorado-compliance program around SB24-205, do not stop. The law is in force. Enforcement begins February 1, 2027, and the case will not resolve before then. But you should be doing three things this week.

Document why your impact assessments exist independently of Colorado. Tie them to Title VII for employment models, ECOA for credit, the Fair Housing Act for housing. If Colorado falls, you still need the framework.

Watch the carveout. Any internal program that explicitly references “diversity” or “historic discrimination” as the rationale for a model design choice is now exposed to federal civil rights scrutiny. The DOJ is signaling it will treat affirmative-action-style algorithmic logic as actionable discrimination. That is a sea change for HR tech, lending, and admissions vendors.

Reread your vendor contracts. Most AI vendor agreements have a “comply with applicable law” clause. If the applicable law in Colorado is enjoined or struck down, what happens to the bias testing your vendor contractually owes you? Probably nothing if you wrote it well. A lot, if you let the vendor define “applicable law” narrowly.

What to Do Now

Pull your Colorado SB24-205 readiness memo from the shelf, if you have one, and add a section on Equal Protection exposure for any “diversity expansion” features in your hiring or lending models. If you do not have a readiness memo, this is the week to start one.

For GCs at companies operating in multiple states, the Colorado fight is a preview. The DOJ has now publicly committed to challenging state AI bias laws on Equal Protection grounds. Connecticut’s bill is next on the radar. New York’s frontier model framework was just revised on April 1 to align with federal positions. The states that have not yet passed their AI bias laws will think twice before drafting carveouts that mirror Colorado’s.

The fastest-moving piece is the lawsuit itself. xAI filed April 9. DOJ intervened April 24. Briefing will accelerate. If you advise clients on AI compliance, set a calendar reminder for the first preliminary injunction ruling. That is when you will know whether Colorado’s law survives the summer.