The AI Employment Law Patchwork: How HR Survives a Federal Pullback and Four State Rulebooks
Bindingness: Binding Law · Scope: Multi-state · Employment
Part of AI Regulation Tracker, our running brief on the laws reshaping AI at work.
On December 11, 2025, the White House signed an executive order titled "Ensuring a National Policy Framework for Artificial Intelligence." It directs the Attorney General to stand up a task force to challenge state AI laws and tells the Commerce Department to flag state rules it considers burdensome. It builds on an April 2025 order that already instructed federal agencies to step back from disparate-impact enforcement. The message from Washington is clear: the federal government wants one national rulebook for AI, and it would prefer the states get out of the way.
The problem for HR is that the states have not gotten out of the way. Illinois, California, and Colorado all have AI employment laws on the books or in force right now, and employment-related state AI rules are not among the carve-outs the December order singles out. So you are caught in the gap: a federal government pulling back from enforcement, and a handful of states pushing forward with binding obligations, while litigation over who wins plays out over months or years. This briefing is about how to operate sensibly in that gap instead of freezing until it resolves.
Key takeaways
- The December 11, 2025 federal EO aims to challenge and preempt state AI laws, but it does not erase them. Until a court or Congress says otherwise, state employment rules remain in force.
- Illinois HB 3773, California's FEHA automated-decision rules, and Colorado's AI Act all impose real obligations on employers today, regardless of the federal posture.
- The safe operating posture is to comply with the strictest state you touch. A program built to the high-water mark satisfies the others and survives whichever way preemption breaks.
- A federal pullback on disparate-impact enforcement does not remove your private-lawsuit exposure. Plaintiffs and state agencies can still bring claims under state law.
What the December 2025 order actually does
Read the order for what it changes and what it does not. It is a statement of federal policy and a set of instructions to federal agencies. It tells the Attorney General to form a task force to identify and challenge state AI laws the administration views as unconstitutional or preempted, and it tells Commerce to catalog state rules that burden interstate commerce. That is real, and over time it could reshape the landscape through litigation and possible federal legislation.
What it does not do is repeal anything. An executive order cannot strike down a duly enacted state statute on its own. Preemption is decided by courts, and challenging a law is not the same as winning. So on the day you read this, Illinois HB 3773 is still in force, California's automated-decision rules under FEHA are still in their first compliance year, and Colorado's AI Act is still on its timeline. The federal order signals where the fight is heading. It does not change the rules you are bound by today.
There is a second, quieter piece that matters more for HR than the headline. The April 2025 order told agencies like the EEOC to deprioritize disparate-impact theories of discrimination. That lowers the odds of a federal enforcement action built on statistical disparity. It does nothing to your exposure under state law, and nothing to a private plaintiff's ability to sue. A reduced federal appetite for enforcement is not a shield. It is one actor stepping back while others stay fully active.
Why HR is caught in the middle
Most functions can watch this fight from a distance. HR cannot, because the highest-risk AI use in any company is a decision about a person, and that is exactly what the state laws govern. Recruiting screens, promotion tools, discipline and discharge recommendations: these are the uses Illinois, California, and Colorado wrote their rules around. When the federal government pulls back and the states press forward, the friction lands squarely on the desk of whoever owns hiring and people decisions.
A federal government that stops enforcing is not the same as a law that stops applying. HR pays for that distinction if it confuses the two.
The trap is to read the federal posture as permission. A People leader who hears "the administration is challenging these state laws" and quietly relaxes the company's controls is making a bet that the challenge will succeed, that it will succeed quickly, and that no state agency or plaintiff acts in the meantime. That is three bets stacked on top of each other, with a discrimination claim as the downside. The far cheaper position is to keep your controls in place and let the lawyers fight about preemption without your compliance program riding on the outcome.
The strictest-state operating posture
The clean way through a patchwork is to stop trying to comply with each rule separately and instead build to the strictest requirement you face anywhere you operate. If your highest-bar state demands written AI-use notice, four-year recordkeeping, and a documented bias check, build all three into your standard process and run them everywhere. A program built to the high-water mark is automatically compliant with the lower bars, and it does not have to be rebuilt if preemption fails or a new state joins the list.
| Pressure | What it signals | What HR should do |
|---|---|---|
| Federal EO (Dec 2025) | Wants one national rulebook; challenging state laws | Track it; do not relax controls on the strength of a challenge that has not won |
| Federal enforcement pullback (Apr 2025) | EEOC deprioritizing disparate impact | Note that private suits and state claims are unaffected; keep your bias checks |
| State laws (IL, CA, CO) | Binding now; not in the EO carve-outs | Comply with the strictest one you touch and run that standard everywhere |
| Litigation timeline | Preemption decided by courts over months or years | Operate as if today's rules hold, because they do until a court says otherwise |
This is also where an AI assistant earns its place. The hardest part of a multistate posture is keeping a current, accurate map of which obligations apply where, and translating that into one internal standard your managers can actually follow. Claude can help you build a plain-language compliance matrix that lists each state you operate in, the AI-employment obligation it imposes, and the single strictest version you will adopt company-wide, then keep that matrix updated as the litigation and the state rules move. The tool drafts and organizes faster than a person working by hand. A named human in HR or Legal still confirms each entry against current guidance and owns the result.
What to do this quarter
You do not need the preemption fight to settle before you act. Build the durable program now.
- Map every state where you have headcount and list the AI-employment obligation each one imposes today. You cannot comply with the strictest state until you know which states you touch.
- Adopt one company-wide standard set to the high-water mark: written AI-use notice, a documented bias check on every AI-touched selection step, and a multi-year recordkeeping log.
- Keep your bias checks even where federal enforcement has eased, because state agencies and private plaintiffs are unaffected by the federal pullback.
- Name a human decision-maker of record for every covered employment decision, so you can show a person owned the outcome regardless of which rulebook ends up applying.
- Set a quarterly review to refresh the matrix as the federal challenge, the state rules, and your own footprint change.
The individual state pieces in this series go deeper on each rulebook. Illinois HB 3773 covers the live notice law whose rules were just pulled, California's FEHA automated-decision rules cover the bias-audit-as-defense framing, and Colorado's AI hiring law covers what HR owes there. Solve the strictest of the three cleanly and you have built most of what a fourth state will demand.
Guardrails
A few lines keep this from going wrong. Do not treat the federal order as a repeal; it challenges state laws but does not erase them. Do not read the enforcement pullback as immunity; private suits and state agency actions are untouched. Do not build a separate compliance process per state when one strictest-state standard covers them all more cheaply. And do not let an AI assistant produce a compliance matrix that a human then relies on without checking, because the rules are moving fast and a stale or hallucinated entry is worse than none. Use the tool to draft and maintain the map. Keep the legal judgment with counsel and a named owner in HR.
What AI does not replace
AI can build and maintain your multistate matrix, draft the company-wide standard, and keep the recordkeeping log consistent across locations. It cannot decide how preemption will break, it cannot tell you which state's law ultimately governs a specific decision, and it cannot carry the liability if your process produces a discriminatory result. Those judgments belong to your counsel and to the human who owns people decisions. The assistant makes the operating posture faster to build and easier to keep current. The accountability stays human, which is the whole point of building these skills deliberately. That is the premise of The Leveraged HR Professional, our course on building AI-assisted HR systems that hold up when the rules are in motion.
Frequently asked questions
Does the December 2025 federal AI executive order cancel state AI employment laws?
No. The order directs federal agencies to challenge and try to preempt state AI laws, but an executive order cannot strike down a state statute on its own. Preemption is decided by courts. Until a court rules or Congress acts, Illinois HB 3773, California's FEHA automated-decision rules, and Colorado's AI Act remain in force, and employment-related state rules are not in the order's carve-outs.
If the EEOC is deprioritizing disparate impact, can I relax our bias checks?
No. The federal enforcement pullback lowers the odds of a federal action built on statistical disparity, but it does nothing to your exposure under state law and nothing to a private plaintiff's ability to sue. State agencies and individuals can still bring discrimination claims. Keeping a documented bias check on every AI-touched selection step is the cheaper position by far.
How should a multistate employer comply while preemption is unsettled?
Comply with the strictest state you touch and run that standard everywhere. Build one company-wide program set to the high-water mark: written AI-use notice, a documented bias check, and multi-year recordkeeping. A program built to the strictest requirement is automatically compliant with the lower bars and does not need to be rebuilt if the federal challenge fails or a new state joins the list.
Related briefings: Illinois HB 3773 AI employment notice · California FEHA AI bias-audit defense · Colorado AI hiring law: what HR owes now