Most of what HR teams do with AI is writing: job ads, policy drafts, interview questions, answers to the same benefits question for the fortieth time. Most AI hiring rules leave that work alone, though discrimination law, including Illinois' AI ban, still applies to what the drafts say. The line that matters sits somewhere else, and many teams cross it without noticing: the moment a tool ranks applicants, scores an interview or flags who to promote, it's making part of a decision about a person. That's the use hiring laws in New York City, Illinois, California, Colorado and Connecticut were written for, and it's often a feature switched on inside software you already pay for.
What is AI in HR?
AI in HR is the use of AI tools by a human resources team, either to help a person write and answer (drafting job ads, policies and replies, summarizing notes) or to shape decisions about candidates and employees (screening, ranking, scoring, flagging), and the second kind is the one hiring laws regulate.
This guide is for HR and People leads at companies of 20 to 1,000 people, and for the COOs and IT leads who share the rollout with them. It covers where AI earns its keep in HR, which uses the US hiring laws reach as of November 2026, and the five things to set up first. It isn't legal advice.
How is AI used in HR today?
SHRM's State of AI in HR 2026, published in March 2026, surveyed HR professionals in December 2025. Of the 1,908 in its sample, 39% said AI was already adopted in their HR function and 7% planned to launch it in 2026. Size matters: 33% of small organizations (2 to 99 employees) and 35% of midsize ones (100 to 499) had AI in HR, against 60% of organizations with 5,000 or more.
Where HR does use AI, SHRM found it most often in recruiting (27%), then HR technology (21%), learning and development (17%) and employee experience (14%). Recruiting is also where the law is. That's the gap this guide is about.
The useful split isn't by HR function. It's by what the output does. In the first lane, AI helps a person write: the person reads the draft, changes it and owns it. In the second, AI's output shapes a decision about a person, often before anyone has looked at the person at all.
The same tool can sit in either lane. A chatbot that drafts a rejection email is in the first. The same chatbot, asked which ten of 200 résumés deserve a call, is in the second. Your applicant tracking system is in the second the day someone switches on its "match score".
How to use AI in HR where it helps most
These are the uses where AI saves real time and a person stays the author.
| Task | What AI does | What a person still does |
|---|---|---|
| Job ads and descriptions | Drafts from a role brief, rewrites for clarity | Checks requirements are real, removes needless ones |
| Policies and handbooks | Drafts sections, compares against a template | Checks against law and practice, approves |
| Staff questions | Answers from your own policy documents | Owns the source documents, handles anything sensitive |
| Interview prep | Drafts structured questions and scoring guides | Agrees the criteria before seeing candidates |
| Notes and summaries | Summarizes interview or meeting notes | Reads the original where it matters |
| Onboarding and training | Drafts checklists, explainers and quizzes | Keeps them current |
Two rules keep this lane safe. First, use a business plan whose terms don't let the vendor train on your data, because HR text is full of personal data (our comparison of ChatGPT Plus and Business shows what changes between plans). Second, have a written output standard: whoever sends the draft owns it. That's the rule in the output standard lesson of our free all-staff course.
Which uses do the hiring laws reach?
The hiring laws reach AI whose output shapes a decision about a candidate or employee: a score, a ranking, a recommendation or a classification that someone sees before deciding, or that decides who they see. Drafting, summarizing and scheduling fall outside most of them. Illinois is the exception to watch: its law defines AI to include generative AI and bans any AI use with a discriminatory effect in recruitment, training or the terms of employment, so a biased AI-drafted job ad is in scope even though nothing was ranked.
As of February 2026, SHRM counted 19 of the most populous states with laws or regulations on employers' use of AI, and 57% of HR professionals working in those states said they weren't aware of them. Here are the rules that matter most to a US employer in November 2026.
| Where | Starts | What it reaches | What you must do |
|---|---|---|---|
| New York City (Local Law 144) | Enforced since July 5, 2023 | Tools that "substantially assist or replace" discretionary decisions to screen candidates, or employees for promotion, for jobs in the city | A bias audit no more than one year before use, a public summary of the results, and notice to candidates and employees who live in the city at least 10 business days before use |
| California (Civil Rights Council rules) | October 1, 2025 | Automated-decision systems used by employers with 5 or more employees, or by their agents | Keep automated-decision system data with your other records for four years. Anti-bias testing, or the lack of it, is relevant to a discrimination claim or defense |
| Illinois (HB 3773) | January 1, 2026 | AI used in recruitment, hiring, promotion, training, discipline, discharge or the terms of employment | No AI that has the effect of discriminating on a protected basis, no zip codes as a proxy, and notice to employees when AI is used for these decisions |
| Texas (HB 149) | January 1, 2026 | AI deployed with intent to discriminate | Don't deploy AI intending to discriminate. Disparate impact alone doesn't show intent |
| Connecticut (Public Act 26-15) | October 1, 2026 | Automated employment-related decision technology | Using one "shall not be a defense" to a discrimination claim. From October 1, 2027, written notice before an employment-related decision is made with one |
| Colorado (SB 26-189) | January 1, 2027 | Technology that materially influences a decision about employees or Colorado-resident applicants | Notice before use, an explanation within 30 days of an adverse outcome, a way to correct data and ask for human review, and three years of records |
California added a second law in September 2026. SB 947, the No Robo Bosses Act, signed on September 30, 2026, bars relying only on an automated-decision system to discipline or fire someone and requires human oversight and notice to workers. According to reporting by HR Dive, its duties start on July 1, 2027.
Two definitions show where the line sits. Colorado's law excludes tools used "solely to summarize, organize, translate, draft, route, or present information for human review", as long as the system doesn't produce "a score, ranking, recommendation, classification, prediction, or other inference" that materially influences the outcome. Connecticut's definition of an employment-related decision leaves out scheduling and productivity monitoring. Drafting is in the first lane. Ranking is in the second. In Illinois, and under discrimination law everywhere, what a draft says can still be the problem, which is why the first lane needs a person who owns every word. Our state guides cover each law in full: New York City, Illinois, Texas, Connecticut and Colorado.
Federal discrimination law doesn't step aside either, and it can reach the vendor too. In Mobley v. Workday, a federal court in California granted preliminary certification in May 2025 of a nationwide collective action under the Age Discrimination in Employment Act, on behalf of applicants aged 40 and over who were denied employment recommendations through Workday's platform. As of November 2026, the case was still in discovery, with no ruling on the merits.
Who answers when an AI hiring tool gets it wrong?
You do, more than you'd expect. The laws above put the duties on the employer that uses the tool, not only on the vendor that built it. New York City makes it unlawful for the employer to use a tool without a bias audit. Illinois puts the notice duty and the discriminatory-effect ban on the employer. Connecticut has made the use of an automated tool no defense to a discrimination claim, and California counts your anti-bias testing, or the lack of it, when a claim is made. Colorado's duties fall on whoever uses the tool to decide.
That changes how you buy. A vendor who can't tell you how its tool reaches a score has handed you a decision you can't explain. Ask before you switch anything on.
A reviewer has to be able to say no
Every one of these laws assumes a person is accountable for the decision. Colorado's spells out what that means: a person "who has authority to approve, modify, or override" the decision, who considers the relevant evidence, is trained to conduct the review and "does not default to the system output". A recruiter who approves a ranked list without looking past it isn't that reviewer.
Our framework holds every high-stakes use of AI to the same standard: the checker needs the time, the information and the authority to reject the output. Shared and stakes, in module 6 of our free course for leaders, uses a recruiter ranking résumés as its example of a high-stakes use that nobody decided to build, and shows how to name the check for it. For HR, that check is a job: someone whose calendar has room for it.
What to set up first
Five steps, in order. Each one produces what the next one needs.
01 · List every HR tool that scores, ranks or screens people
Start with your applicant tracking system's matching and ranking features, assessment and video-interview platforms, background-screening scores and performance software. Then ask recruiters and managers what they use chatbots for. The free Shadow AI Audit covers AI features switched on inside software you already pay for. Your AI list, in module 3 of our course, turns what you find into a list with an owner on every row.
02 · Put each use in a lane, and write down why
For each tool and each use, ask one question: does its output shape a decision about a candidate or employee? If yes, it's in the decision lane, and you check it against the laws for every place you hire. If it only drafts, summarizes or schedules, write down why, so you can show your reasoning later.
03 · Write the HR section of your AI policy
Say which tools may hold employee and applicant data, that chatbot output is never used to decide who gets hired, promoted or let go, and who signs off before any scoring feature is switched on. The AI Usage Policy generator drafts the policy, and our AI policy template has clauses you can copy.
04 · Get the paperwork for every decision-lane tool
For each tool in the decision lane, get from the vendor what the laws in your states require: a bias audit if you hire in New York City, how scores are produced, what data the tool uses and its known limits. Draft the notices candidates and staff must get, and check your contracts for indemnities. The state guides linked above list each law's duties.
05 · Name a reviewer and an owner for each one
Give each decision-lane tool an owner, a backup and a review date, and name who reviews a contested decision. Train them on what the tool does and doesn't consider. Keep records of notices, requests and reviews. Owners and backups, in module 7 of our course, sets this up. If you'd rather have help with the whole rollout, that's what our AI implementation work does.
Done in this order, most of the work lands in the first lane, where AI saves time and the risk is a data rule. The decision lane stays small, documented and owned. That's what deploying AI well in HR looks like.
What's still open
- Colorado's rules. The attorney general held the public hearing on draft rules on October 26, 2026. As of this update no final rules have been adopted, and the law starts on January 1, 2027.
- California's SB 947. We've taken its July 2027 start date from secondary reporting; we'll confirm it against the chaptered text.
- Mobley v. Workday. The plaintiffs have asked the court to certify a class, with a hearing set for March 9, 2027. A ruling would be the first big test of whether an AI hiring vendor answers for its screening.
We'll update this article when these move. The date at the top of the page shows when we last checked.
Sources
- Int 1894-2020, Local Law 144 of 2021, New York City Council, 2021: the law's definitions, bias audit, notice and penalties.
- Automated Employment Decision Tools: Frequently Asked Questions and Automated Employment Decision Tools, NYC Department of Consumer and Worker Protection: scope and the July 5, 2023, enforcement date.
- 775 ILCS 5/2-102 and 775 ILCS 5/2-101, Illinois General Assembly: the AI ban, the notice duty and the definition of employer.
- Automated-decision systems regulations, final text and rulemaking actions, California Civil Rights Council, 2025: the October 1, 2025, start, four-year records and anti-bias testing.
- Governor Newsom signs more first-in-the-nation worker protections, Office of the Governor of California, September 2026, and Newsom signs McNerney's No Robo Bosses Act of 2026, Office of Senator McNerney, September 2026: SB 947's signing and requirements.
- California revamps AI protections for workers, HR Dive, October 2026 (secondary): SB 947's July 1, 2027, start date.
- House Bill 149, enrolled version, Texas Legislature, 2025: the intent rule and the January 1, 2026, effective date.
- Public Act 26-15, Connecticut General Assembly, May 2026: the no-defense rule, the 2027 notice and the definition of employment-related decision.
- SB 26-189 and the signed act, Colorado General Assembly, May 2026: the duties, the summarize-and-draft exclusion, meaningful human review and the January 1, 2027, start.
- Automated Decision-Making Technology rulemaking, Colorado Attorney General, November 2026: the status of the rules.
- The State of AI in HR 2026 and the full report, SHRM, March 2026: adoption by size, practice areas and awareness of state AI laws.
- Mobley v. Workday docket, CourtListener, November 2026: the case's status and the March 9, 2027, hearing.
- Federal Court Allows Collective Action Lawsuit Over Alleged AI Hiring Bias, Holland & Knight, May 2025 (secondary): the preliminary certification.