AI hiring, in plain English
Is AI résumé screening legal?
Short answer: yes, almost everywhere — surveys suggest the large majority of employers already use some automation in hiring. But it is regulated, the rules are tightening through 2026, and they all converge on the same three duties: test for bias, keep a human making the real decisions, and be able to explain every score. Do those three and AI screening is defensible. Skip them and the risk isn’t the AI — the exposure lands on you, not your vendor.
How screening AI actually goes wrong
The famous cautionary tale: a major tech company trained a résumé screener on ten years of its own hiring decisions. Because those past hires skewed male, the model learned to downrank résumés that signaled “woman” — even penalizing the word “women’s” — and had to be scrapped. Nobody wrote a biased rule. The tool just faithfully learned biased history.
That is the core failure mode in three forms. Training on your past hires bakes yesterday’s bias into tomorrow’s shortlist. Proxies do the same thing sideways: zip codes, school names, and employment gaps can quietly stand in for race, class, age, or parenthood. And black-box scoring makes it all undiagnosable — if nobody can say why a candidate scored 42, nobody can prove the 42 was fair, including you. The pattern is live enough that a major HR-software vendor is currently defending a class action over alleged algorithmic discrimination in exactly this kind of screening.
The rules to know in 2026
You don’t need a law degree — the map is short, and every entry points the same direction:
US federal: the EEOC and Title VII
There is no federal ban on AI screening. Instead, the decades-old discrimination rules apply to the algorithm exactly as they would to a human: if a tool disproportionately filters out a protected group (disparate impact), the employer is liable — and the EEOC has said plainly that you can’t outsource that liability to your software vendor. Your tool, your hire, your responsibility.
New York City: Local Law 144
If an automated tool substantially helps decide who gets hired or promoted for a NYC role, it must get an independent bias audit every year, the audit results must be public, and candidates must be told the tool is in use. It became the de-facto template other regulators borrow from.
Illinois and a growing state patchwork
Illinois regulates AI analysis of video interviews (consent and disclosure) and, from 2026, restricts AI that discriminates in employment decisions. Several other states have proposals in flight. The direction of travel is the same everywhere: disclosure, human oversight, no discriminatory outcomes.
Colorado: the AI Act
Colorado’s law treats hiring tools as “high-risk” AI: developers and employers using them owe a duty of reasonable care to avoid algorithmic discrimination, with risk-management, impact assessments, and notices to candidates. It signals where US state law is heading.
EU: the AI Act
The EU AI Act classifies recruitment and screening systems as high-risk. That brings mandatory transparency, human oversight, logging, and bias controls, with obligations phasing in through August 2026 and serious fines behind them. If you ever hire into the EU, this one is not optional.
The compliant setup, in five habits
Regulation-proofing AI screening isn’t a legal project. For a small team it comes down to five working habits:
- 1
Score only what the job needs. Every criterion the AI scores should trace back to the job description — skills, experience, evidence of the actual work. Nothing else gets scored, so there is nothing else to be biased about.
- 2
Require a reason for every score. If the tool can’t show which part of the résumé produced the number, you can’t check it, can’t correct it, and can’t defend it later. Explainability isn’t a nice-to-have; it’s the difference between a screening system and a liability.
- 3
Keep a human on every real decision. AI compresses the pile and drafts the shortlist; a person decides who interviews and who gets rejected. Fully automated rejection is exactly the pattern regulators (and lawsuits) target.
- 4
Watch your outcomes. Periodically look at who is scoring high and low. If the pattern looks skewed against a group, stop and investigate — that check is essentially what NYC’s bias audit formalizes.
- 5
Disclose and document. Tell candidates when an automated tool is involved where the law requires it, and keep records of criteria and decisions. Boring, cheap, and the first thing anyone asks for if a hire is ever challenged.
The upside nobody mentions
Here is the irony in the fear: the legally risky screening method isn’t AI — it’s the one most small businesses use today. A tired human skimming 200 résumés at 9pm applies no consistent criteria, leaves no record, and couldn’t explain half of their passes the next morning. A tool that scores every applicant against the same job-derived criteria, shows its reasons, and leaves the decision to you is not just faster. Done right, it’s the more defensible process — and the fairer one for the people in the pile.
One honest note: this page is general information, not legal advice. The rules named here are evolving and vary by where you and your candidates are — for specific decisions, talk to an employment lawyer.
Screening you can explain
Hired Copilot was built around the three duties: it scores only against your job description, shows the reasons behind every score, and never auto-rejects anyone — every real decision stays with a human. That’s the version of AI screening you can stand behind.
Keep reading
- How to screen job applicants fast, without a recruiter The five-step first pass for small teams.
- AI résumé screening software What it does, and what it should never do.
- Hired Copilot for employers Explainable screening and structured first-round interviews.