What is adverse impact?

The legal idea that a hiring practice can be illegal even when you had zero intent to discriminate, if it screens out a protected group at a meaningfully higher rate. It is the single biggest reason to keep a human in charge of every AI hiring decision.

You did not set out to discriminate. You set out to handle 300 applicants faster, so you let an AI tool rank them. That is precisely the situation the law cares about, because under federal employment law, intent does not save you. Outcome does the talking, and you are on the hook even if the tool came from a vendor.

What it is (in plain English)

Adverse impact (also called disparate impact) comes from Title VII of the Civil Rights Act. A hiring practice that looks neutral can still be unlawful if it disproportionately screens out people based on race, sex, age, or another protected characteristic, unless you can show the practice is job-related and necessary.

The rough yardstick regulators use is the four-fifths rule: if the selection rate for one group is less than 80% of the rate for the group selected most often, that is a red flag for adverse impact. Example: if 60% of male applicants advance but only 30% of female applicants do, the female rate (30%) is half the male rate, well under the 80% threshold, and you have a problem to explain.

Two facts make this urgent for AI specifically. First, the bias is real and measured: a 2025 study running hundreds of thousands of test resumes through leading models found names associated with white candidates preferred in the large majority of head-to-head matchups against names associated with Black candidates. The model picks up the signal from names, schools, and addresses whether you want it to or not. Second, the law puts the liability on you, the employer, not on the vendor whose tool produced the skew. You cannot outsource the responsibility.

Why CEOs care

Because the rules are no longer theoretical, and "the AI did it" is not a defense.

The compliance landscape is now live, not coming:

• New York City Local Law 144 requires an annual independent bias audit, public posting of the results, and advance notice to candidates for any "automated employment decision tool" (an AEDT), with daily penalties for violations.
• The Colorado AI Act treats AI used in employment decisions as high-risk and takes effect June 30, 2026.
• Illinois amended its civil rights law (HB 3773) to cover AI in employment, effective January 2026.

Here is the practical line that keeps you safe and still lets you move fast. There is a genuine grey area about whether an AI that merely summarizes resumes counts as a regulated AEDT, the laws generally target tools that substantially assist or replace the decision by scoring or ranking. So the safe posture, the one every hiring workflow on this site uses, is this: let AI summarize and organize, never let it score, rank, or auto-reject, and keep a human making the actual call with an audit trail of who decided. Then run a simple four-fifths check on your funnel periodically to catch skew early. And because this is law that varies by state and city, run your specific setup past employment counsel before you scale it. That is not a disclaimer, it is the cheapest insurance you will ever buy.

Where you'll see it

• In every piece of the hiring cluster: screening resumes, interview scorecards, job descriptions.
• The moment your ATS offers an AI feature that scores or ranks candidates rather than just summarizing them.
• Any vendor demo that promises to "rank your best candidates automatically." That is the feature to interrogate, not celebrate.

What to do next

Before you point any AI at a stack of resumes, write down your job-related must-haves first, and decide that AI will summarize against them while you do the ranking. Then put a recurring reminder to run a four-fifths check on your hiring funnel. If you are already using an AI tool that scores candidates, that is the thing to take to counsel this month. Tell me what your hiring stack does today.