NEW AI HIRING LAWS ASK FOR A NOTICE AND A HUMAN. THE HUMAN IS THE HARD PART.

For two years, the argument about AI screening has been about whether it's accurate and who's liable when it isn't. Both matter. Neither is the next thing landing on a talent leader's desk.

On January 1, 2027, Colorado and California join Illinois and New York City in requiring employers to tell candidates when an automated tool is affecting a hiring decision. California goes further and adds an opt-out right, with exceptions. Read the laws side by side, and they ask for the same two things: a notice, and a human who can look again.

The notice is the easy part. The human is the part most AI screening was bought to remove.

A note on where I sit: I am a strategic advisor to RoundOne.ai, which builds AI hiring tools. This piece does not recommend any vendor.

What do AI hiring disclosure laws actually require?

AI disclosure in hiring means telling candidates, before it happens, that an automated tool is evaluating them, what it does, and how they can reach a person. As of October 2026, four jurisdictions matter most for US talent teams:

• New York City. Local Law 144 requires notice at least 10 business days before using an automated employment decision tool, plus a published annual bias audit. It does not guarantee anyone a human review.
• Illinois. Notice has been required since January 1, 2026. The state published draft rules on what a compliant notice looks like, then withdrew them in June. The duty exists. The instructions don't.
• Colorado. Starting January 1, 2027, SB 26-189 requires notice before use, a plain-language explanation within 30 days when the tool materially influenced a rejection, and human review on request where commercially reasonable.
• California. Starting January 1, 2027, businesses covered by the CCPA that use automated decision-making for significant decisions, including hiring, must provide a pre-use notice, answer access requests, and offer an opt-out unless an exception applies.

Whether each law covers you depends on size, where your candidates live, and how meaningfully a person reviews the tool's output. That is a question for counsel. This piece is about the part counsel can't build for you.

If you hire candidates in the EU, the EU AI Act is a separate and heavier set of rules. It classifies AI used in hiring as high-risk.

Why is this a talent leader problem and not a legal one?

Because the laws describe outcomes, and only the hiring process can produce them.

Legal can compose a notice in an afternoon. Legal can't decide who reviews an applicant who opts out, how fast, against what criteria, or what a recruiter says when a rejected candidate asks why. Those are process decisions. The process belongs to talent.

Most teams haven't noticed it's theirs. As of February 2026, SHRM found that 19 of the most populous states had enacted AI rules affecting employment decisions, and 57% of HR professionals in those states didn't know. That isn't a training gap. It's an ownership gap. When a requirement sits between legal, the vendor, and TA, each one assumes another has it.

Here's the uncomfortable part. Many AI screening tools were purchased on the promise that recruiters would stop reading the top of the funnel. Every one of these laws assumes a person is still reachable. If that promise was the business case for your tool, the law just reopened the line item.

Isn't a careers-page notice enough?

It may satisfy a rule. It won't satisfy the candidate, and in practice, the candidate enforces it.

Candidates already assume software is involved. Gartner found 52% believe AI screens their application, and only 26% trust it to evaluate them fairly. What they're missing is information. Greenhouse's 2026 survey of 2,950 job seekers across five countries found 70% weren't told AI was involved, and only 18% say employers have clear AI policies.

A notice buried on a careers page closes that gap on paper. A notice without a fallback is a disclaimer. New York City shows what that's worth. When the New York State Comptroller audited enforcement of Local Law 144, it found the city had received two complaints in two years and never checked whether its intake process worked. Enforcement there runs on complaints, which means it runs on candidates noticing. Colorado and California build in the same mechanic: the candidate asks, and you have to answer.

So the test of your notice isn't whether legal approved it. It's whether a candidate who reads it knows what happens next, and whether that thing actually happens.

Who reads the opt-out queue?

If you can't name a person, nobody does.

An opt-out or appeal path isn't a checkbox. It's a second hiring process running beside the first, without the automation that made the first one fast. We've covered the candidate-side opt-out. From the employer's side, five questions decide whether the path is real:

• Who owns it? A named role, not "the team."
• How fast? If the automated path returns a decision in two days and the human path takes three weeks, opting out is a penalty.
• Against what? A reviewer needs the same criteria the tool was configured on, in writing. If nobody can write them down, that's its own finding.
• In what order? Applicants reviewed after the shortlist is already full were rejected by the calendar.
• How would you know? Track volume, time to review, and how often the human path advances someone compared with the automated one. A path that never advances anyone isn't a path.

California makes the burden explicit. For hiring, one way an employer can decline an opt-out is by making sure the tool works as intended and doesn't discriminate. And where a person is in the loop, the regulations expect that person to know how to interpret the tool's output and to have the authority to change the decision. Colorado requires an explanation of the tool's role within 30 days of a rejection it influenced. Both put the same burden on you: understand what the tool is actually deciding. If your vendor can't explain why one candidate was screened out, neither can you.

What should talent leaders check before January?

Six things, none of which require a new tool:

• List every tool that scores, ranks, filters, or interviews candidates, including features quietly switched on inside your ATS.
• Map where your candidates live, not where your offices are. Several of these laws follow the candidate.
• Read your current notice as a candidate would. Does it say what happens if they ask for a person?
• Name the owner of the human path and give it a response time.
• Ask your vendor what explanation it can produce for a single rejected candidate, in plain language.
• Take the coverage questions to counsel with this list in hand.

So what are these laws really asking for?

These laws aren't asking you to stop using AI. They're asking you to prove there's still a hiring process behind it.

Teams that built the screen and skipped the fallback will end up retrofitting both, probably in response to a candidate who asked a question nobody could answer. Teams that treat the notice as a promise, and staff the promise, will have something most candidates say they've never seen: a process that tells them the truth and lets them reach a person.

This piece shows the rules as of October 2026 and isn't legal advice.

Because good enough, isn't.