Federal and State Law Are Pulling Apart on AI Hiring Bias — What Employers Should Do
NEW YORK | June 15, 2026 — If your company uses software to screen resumes, rank candidates, scan video interviews, or flag people for promotion, you are now operating under two sets of rules that are actively moving in opposite directions. It happened quietly, without a single headline moment, which is part of why so many employers have missed it. But it's the reality as of mid-2026, and it's worth understanding before your next hiring cycle.
For most of the last decade, employment law had a workable center of gravity: a shared federal understanding of what counted as discrimination, with states adding their own layers on top. That center is gone. State and federal law are wrestling with the same question: how much an employer should answer for a hiring practice that produces discriminatory results, even without clear evidence of an intent to discriminate. The federal government has taken the position that employers should not be liable at all in such circumstances, retreating from the decades-old "disparate impact" doctrine (more on what that is in a moment). Federal authorities now suggest that some of the very measures employers adopted to prevent disparate impact — proactive bias testing, race-conscious balancing — may themselves be unlawful. Conversely, a growing group of states is doubling down, writing detailed new rules that hold employers responsible for biased outcomes from the software that screens, scores, and ranks job candidates. The result is a genuine split, and it has a practical consequence most businesses haven't absorbed yet: a single national hiring process may no longer be something you can safely run.
Here's what's actually happening, what it means for your business, and what to do about it.
A fifty-year-old doctrine meets modern hiring
To see why the current split matters, employers must understand the doctrine sitting at the center of it: disparate impact.
For most of employment law's history, proving discrimination meant proving intent: showing that an employer meant to treat someone worse because of their race, sex, or age. In 1971, the Supreme Court changed that. In Griggs v. Duke Power, the company required a high-school diploma and a passing aptitude-test score for certain jobs, requirements that screened out Black applicants at far higher rates. The Court held that a practice neutral on its face could still violate Title VII even with no intent to discriminate. The employer's only defense was to show the practice was actually job-related and necessary for the business. Effect, not just motive, could be unlawful.
That idea became one of the most consequential in American employment law. For two decades, the Griggs decision served as the baseline for how employers had to navigate the practical effect of their employment and hiring policies. Then, in 1991, Congress wrote the Griggs analysis directly into Title VII. For thirty-five years since, disparate impact has been the rule that quietly governs the tools employers use to screen people: the tests, the cutoffs, the scoring systems. The legal question was never just "did you mean to discriminate," but "does this screen push out protected groups at higher rates without a real business justification."
For most of that history, the "tools" were straightforward: a test score, a credential, a physical requirement. What changed is that the tools got intelligent. Employers now hand the screening, ranking, and scoring to software (resume parsers, video-interview analyzers, candidate-matching algorithms, promotion models) that makes decisions at a scale and opacity no diploma requirement ever did. An algorithm trained on a company's past hires can quietly reproduce that company's past patterns, producing exactly the kind of facially neutral, statistically skewed outcome Griggs was built to reach. Disparate impact, a doctrine older than the internet, turned out to be the main legal check on the most modern hiring technology there is.
Which is why it matters so much that state and federal law are now treating that doctrine in opposite ways.
The federal retreat from disparate impact
In April 2025, an executive order directed federal agencies to step back from disparate-impact liability. The Equal Employment Opportunity Commission then withdrew its earlier guidance on AI-driven selection tools — removing the 2023 technical-assistance document from its website in January 2025 — and shifted its enforcement focus toward intentional discrimination.
The most striking move came on June 9, 2026, when the Department of Justice's Office of Legal Counsel issued an opinion concluding that the EEOC's disparate-impact guidelines are unconstitutional. Its reasoning: holding employers liable for purely statistical disparities effectively pressures them into race-conscious decisions.
It's worth being precise about what that opinion is, because the headlines blur it. It is a government legal memo, not a court ruling. Disparate-impact liability has not been struck down by the courts; it remains part of Title VII, and private plaintiffs can still bring these claims. What has shifted is the federal enforcement posture: the agencies that once brought these cases are stepping back. The doctrine is still on the books. But the federal government has made clear it will no longer pursue these claims. That distinction matters enormously for how employers should plan.
The Department of Justice's opinion does not just take the position that statistical disparities, on their own, are not evidence of discrimination. It argues that trying to correct those disparities may itself be a form of discrimination. The proactive steps the long-standing federal benchmarks encouraged — validating selection tools across demographic groups, in line with the 1978 Uniform Guidelines on Employee Selection Procedures, or adjusting a process to balance outcomes by race under the EEOC's affirmative-action regulations — are now cast as the constitutional problem rather than the solution. An employer who built a careful, race-conscious bias-prevention program to stay on the right side of the law could now find that same program drawing federal scrutiny. The federal posture isn't simply "we'll stop watching." For some practices, it's closer to "the thing you did to comply is the thing we object to."
The state advance on AI tools
While Washington pulls back, several states are building exactly the kind of guardrails the federal government is stepping away from — aimed squarely at AI hiring tools.
California is the most demanding. As of October 1, 2025, regulations under the Fair Employment and Housing Act make clear that using an automated decision system can violate California law if it produces a discriminatory impact. Notably, the rules reach the vendors who build these tools, not just the employers who use them. They require proactive bias testing and four years of recordkeeping.
Illinois, effective January 1, 2026, amended its Human Rights Act to prohibit AI that produces discriminatory effects in employment, to require that employers tell applicants when AI is being used, and to ban certain “intrusive” tools like emotion- and attention-detection software.
Colorado is the cautionary tale about how fast this ground moves. Its original 2024 AI law was repealed and replaced in May 2026 by a streamlined version (SB 26-189), now set to take effect January 1, 2027, centered on transparency and a right to human review of adverse decisions.
And New York City has had its Local Law 144 in force since 2023, requiring annual independent bias audits of automated hiring tools, public posting of the results, and advance notice to candidates. After a critical 2025 state audit of its enforcement to date, the city is signaling that it intends to enforce far more actively going forward.
Notice the through-line: these laws don't just ban bias in the abstract. They impose operational duties (testing, disclosure, recordkeeping, human review) that exist whether or not anyone ever files a discrimination claim.
Where this leaves employers
Taken together, the two trends create a genuine dilemma for employers. The federal floor is dropping at the same moment several states are raising their own ceilings. For an employer operating in just one of those states, that's manageable. For a multi-state employer, it creates a real bind: the streamlined approach that's now defensible under federal enforcement is not enough to satisfy California, Illinois, or New York City. Trying to run one uniform national process means either applying the strictest state's rule across every jurisdiction, or quietly falling out of compliance in the ones that demand more.
That dilemma is compounded by the federal government's posture toward disparate-impact analysis as a practice. The bias testing and disclosure that states like California and Illinois now require are precisely the kind of proactive, group-conscious measures the federal government has begun to question. Satisfy the states, and you may draw federal scrutiny; conform to the federal posture, and you may fail to meet the states' requirements. The same hiring tool can pull you in opposite directions at once.
There's a further pressure most employers haven't priced in: even where federal agencies have stepped back, private lawsuits are very much alive. The most-watched example, Mobley v. Workday, is a federal case in California where the court has allowed a novel theory to proceed: that an AI hiring vendor can be directly liable for discrimination as an "agent" of the employers using it. As of a March 2026 ruling, the case has only escalated. The court confirmed that job applicants, not just employees, are protected under federal age-discrimination law, and the age claim is proceeding as a collective action. The "it's just a neutral tool the vendor sold us" defense is weakening, which means the contract you signed with your hiring-software provider deserves a fresh look.
What employers should do
None of this is a reason to rip out your hiring technology or to panic. It is a reason to stop treating "we use a well-known vendor" as the end of the analysis. A few grounded steps:
Know what you're running. Most employers can't actually list every tool in their hiring pipeline that scores, ranks, or filters people. Build that inventory first; you can't manage exposure you haven't mapped.
Match your process to your map. If you hire in California, Illinois, or New York City, the question isn't whether federal enforcement has relaxed; it's whether you're meeting those states' specific testing, notice, and recordkeeping duties. For genuinely multi-state hiring, a single national process is increasingly hard to defend; separate, more rigorous tracks for the strict jurisdictions are worth considering.
Re-read your vendor contract. The standard software agreement disclaims responsibility for discriminatory outcomes and caps the vendor's liability at the subscription price. In the current environment, accepting that as-is can leave you holding risk that should be shared. This is the most concrete, highest-leverage thing on the list.
Keep a human in the loop. Several of these laws are converging on the same requirement: a real person, with real authority, who can review and overturn an automated decision when someone appeals it. Building that pathway now positions you for the 2027 deadlines already on the calendar.
The honest summary is this: the law here is unsettled, it's moving quickly, and it's pulling in two directions at once. That uncertainty is exactly why the worst move is to assume the relaxed federal posture is the whole story. The employers who come through this well won't be the ones who guessed which way the politics would break. They'll be the ones who built hiring practices that hold up under whichever set of rules ends up applying to them.
That's the part I help businesses think through: how to operate with confidence while the ground is still shifting. If you're deploying AI anywhere in your hiring process and you're not sure where you stand, that's a conversation worth having before a problem finds you, not after. To assess how these developments affect your business and build a forward-looking compliance strategy, contact Sharpe Counsel.
Zack G. Sharpe, IV is the founder of Sharpe Counsel, advising businesses, executives, athletes, and creators on employment, contracts, and the legal questions that come with growth.
Attorney Advertising. This article is general information about a developing area of law, not legal advice for any specific situation. Prior results do not guarantee a similar outcome.
