US State AI Hiring Laws for Small Business: What Changes on 1 January 2027 — and What to Do Before Then

If you hire with an AI tool — an applicant-tracking system that ranks CVs, a video-interview screener, a résumé filter — two new sets of rules land on 1 January 2027. Most small businesses are covered by at least one of them, and the fix is mostly paperwork: a notice line, a named human reviewer, and records you keep. This is general information, not legal advice.

The short answer

Does any of this apply to a small business? Yes, in most cases — but not in the way the headlines suggest. The law that reaches the most small firms is Illinois’s, which has applied since 1 January 2026 and covers any employer with one or more Illinois employees for 20 or more weeks a year. New York City’s bias-audit rule has applied since July 2023 with no small-employer carve-out. California’s big new hiring rules, by contrast, mostly do not reach a small firm — its privacy rules apply only to businesses above a $26,625,000 revenue threshold. And Colorado’s rewritten law applies to hiring decisions made on or after 1 January 2027, but the state attorney general has said he will not enforce it until rulemaking is finished. The date matters; the panic does not.

Why it matters

Small businesses rarely hire through a dedicated HR team. They hire through software: Indeed, LinkedIn, an applicant-tracking system (ATS) with a built-in AI ranking feature, maybe a video-interview tool. That software now sits inside a patchwork of state and city rules that were written for large employers but reach small ones too — and the rules disagree with each other on what counts, who is covered, and what you must do. The good news is that the same three habits satisfy most of them: tell candidates when AI is involved, keep a human in the decision, and keep the records. The bad news is that buying the software does not transfer the liability. If the tool discriminates, the employer answers for it — a point a federal court made in Mobley v. Workday, where a judge allowed discrimination claims against the software vendor itself to proceed (more on that below).

What changes, and when

Place Law Status What it asks Who it reaches
New York City Local Law 144 Enforced since 5 July 2023 Independent bias audit within 1 year before use; publish a summary; tell candidates 10 business days before use; offer an alternative process. The only bias-audit requirement anywhere in the US. Employers using an automated employment decision tool for NYC roles. No small-employer carve-out.
Illinois HB 3773 (Public Act 103-0804, Human Rights Act amendment) In force since 1 Jan 2026 No AI use with a discriminatory effect on a protected class (intent irrelevant); no ZIP code as a proxy; tell applicants and employees when AI is used in a covered decision. Any employer with one or more Illinois employees for 20 or more calendar weeks in the year of or before the alleged violation (775 ILCS 5/2-101(B)(1)). That is most small firms with an Illinois employee.
Illinois (video) AI Video Interview Act In force since 2020 Notice, explanation and written consent before AI analyses a video interview; delete the data on request. Employers using AI video interviews in Illinois.
California FEHA / Civil Rights Council ADS regulations In force since 1 Oct 2025 Discriminatory automated decision systems are a civil-rights violation; keep ADS records, selection criteria and applicant data at least 4 years; run anti-bias testing. Employers with 5+ employees.
California CCPA ADMT regulations (Article 11) Compliance due 1 Jan 2027 for ADMT already in use Pre-use notice, opt-out or qualifying human appeal; access requests. Risk assessments for pre-2026 processing due 31 Dec 2027. CCPA “businesses” only: revenue over $26,625,000, or personal information of 100,000+ consumers/households, or 50%+ of revenue from selling/sharing personal information. Most small businesses are excluded even if they hire in California.
California SB 947, “No Robo Bosses Act” Effective 1 July 2027 No sole reliance on an automated decision system for disciplinary or termination decisions; human corroboration required. California employers using ADS for discipline or termination.
California SB 951 Effective 1 Jan 2027 Layoff (WARN) notices must identify affected positions, automated job functions and the technology when AI causes the layoff. California employers filing WARN notices.
Texas HB 149, Responsible AI Governance Act (TRAIGA) In force since 1 Jan 2026 No deploying AI with the intent to discriminate against a protected class. Penalties: $10,000–$12,000 per curable violation, $80,000–$200,000 per uncurable violation, $2,000–$40,000 per day for continuing violations. Enforced by the Texas Attorney General only; no private right of action. Anyone who develops or deploys AI in Texas.
Colorado SB 26-189 (rewrote the 2024 Colorado AI Act before it took effect) Applies to decisions made on or after 1 Jan 2027 Clear pre-use notice; within 30 days of an adverse decision, a plain-language explanation plus how to ask for information and get meaningful human review; 3 years of records. Impact assessments, risk-management programmes and the duty of care were removed by the rewrite. Enforcement by the Colorado Attorney General only, up to $20,000 per violation, with a 60-day cure period (until 1 Jan 2030). Employers “doing business in Colorado” using ADMT that materially influences a consequential employment decision. Excludes independent contractors and non-Colorado residents; general-purpose chatbots and tools that only summarise, translate, draft or route for human review are excluded.
Connecticut Public Act 26-15 (SB 5, AI Responsibility & Transparency Act) Parts in force since 1 Oct 2026; notice duties reach technology deployed on or after 1 Oct 2027 Using automated tech is not a defence to a discrimination complaint (documented testing can mitigate); WARN notices must disclose whether layoffs relate to AI. From 1 Oct 2027: written notice before using automated employment-related decision technology as a substantial factor. AG enforces as a deceptive trade practice; no private right of action; violations through 31 Dec 2027 carry a 60-day cure period. Connecticut employers.
New Jersey N.J.A.C. 13:16 (Division on Civil Rights disparate-impact regulations) Effective 15 Dec 2025 AI screening (résumé screeners, scheduling filters, video analysis) that produces a disparate impact is unlawful; tools must provide a mechanism to request reasonable accommodation. Note the contrast with Connecticut: in New Jersey, reasonable steps are not a defence if they do not actually prevent the disparate impact. New Jersey employers.
Federal (all states) Title VII, ADEA, ADA — no federal AI hiring statute In force now Discrimination in hiring is unlawful in every state, with or without an AI-specific statute. The employer is responsible for the outcome of its selection tools even when a vendor built them. Every US employer (Title VII: 15+ employees; ADEA: 20+; ADA: 15+).

Dates can move — Colorado already moved twice. The original 2024 Colorado AI Act was due to take effect in February 2026, then June 2026, before the May 2026 rewrite pushed it to 1 January 2027. Treat the dates above as what the texts say today, not as promises.

Does this even apply to me? The triage

Before reading any state-by-state list, answer three questions.

1. Does your tool even count as a covered tool?

Scores, ranks, sorts, filters or rejects candidates = covered in most jurisdictions. Drafting a job post, summarising a CV for a human to read, scheduling interviews = generally not covered. Colorado expressly excludes tools that only summarise, translate, draft or route where no score, ranking, recommendation, classification or inference is produced. The line is whether the tool decides something about a person, or merely helps a person decide.

2. Which places actually reach you — by where the candidate lives, not where your office sits?

A Colorado applicant triggers Colorado’s rules from 1 January 2027. An Illinois applicant triggers Illinois’s rules now. A New York City applicant triggers Local Law 144. Your headquarters’ state is not the test; the candidate’s location is.

3. What is the split that decides everything for a 1–10 person firm?

Most small businesses are not covered by California’s CCPA ADMT rules, but are covered by Illinois and New York City. That threshold split is the single most useful fact in this article: if you hire in Illinois or New York City, the rules already apply to you. If you only hire in California, the hiring-specific rules mostly do not — but the older FEHA anti-discrimination rules do if you have five or more employees.

A worked example: a six-person firm hiring one assistant

Copy this flow. A six-person firm hires one assistant. Applicants come via Indeed, are screened by the ATS’s built-in AI ranking, and the job post was drafted in ChatGPT.

Step 1 — inventory what each tool does. The ATS ranking scores candidates: covered. The ChatGPT job-post draft: not covered. Indeed’s own ranking: ask Indeed what it does — it is a third-party tool in the chain, and you are responsible for the outcome either way.

Step 2 — map by candidate location, not HQ. A Colorado applicant triggers SB 26-189 from 1 January 2027. An Illinois applicant triggers HB 3773 now. A New York City applicant triggers Local Law 144 — and that one means asking your ATS vendor for its bias-audit summary, not commissioning your own audit.

Step 3 — the three things to do before 1 January 2027. (1) Add one plain-language notice line to the job post and the application flow. (2) Name a human reviewer for every rejection the tool influences. (3) Keep the criteria, the tool’s output and who made the call — for the retention period that fits the strictest place you hire (California FEHA: four years; Colorado: three years).

Step 4 — five questions to email your ATS vendor. What does the tool score, rank or reject? What data does it use — and does it use ZIP code or similar proxies? Has it been tested for bias, and can we see the summary? Who is the human contact for a candidate appeal? Will you put the notice wording in writing?

Colorado, explained: the law rivals keep getting wrong

Colorado is where most guides go wrong, so it is worth a paragraph of its own. The 2024 Colorado AI Act (SB 24-205) never took effect. It was rewritten in May 2026 by SB 26-189, signed 14 May 2026, and now applies to decisions made on or after 1 January 2027. The rewrite removed the impact assessments, the risk-management programmes and the duty of care that made the original law famous. What remains is narrower: notice, explanation, human review, records.

Two further wrinkles. First, the technology company x.AI sued Colorado in April 2026, and on 27 April 2026 a court stayed enforcement of the old act while the case proceeds. The Colorado attorney general has separately said he will not enforce the new law until the required rulemaking concludes — and that rulemaking has not formally begun (proposed rules were filed 11 August 2026; a public hearing was held 26 October 2026; rules are due by 1 January 2027). So enforcement realistically cannot start before 1 January 2027. The statute is still on the books and the date stands: prepare, don’t panic. Second, a December 2025 federal executive order does not pre-empt any state law by itself. No state AI hiring law has been struck down.

The vendor-liability case: Mobley v. Workday

The case every hiring-tool vendor watches is Mobley v. Workday, filed in February 2023 in the Northern District of California (No. 3:23-cv-00770-RFL). On 12 July 2024, Judge Rita Lin held it plausible that Workday acted as its employer-customers’ “agent” under Title VII, the ADEA and the ADA, so disparate-impact claims went forward; employment-agency and intentional-discrimination theories were dismissed. On 16 May 2025 the court granted preliminary collective certification on the ADEA (age 40+) claim only; roughly 15,000 people had filed consents to join as of 7 October 2026. A class certification hearing on the race, sex, age and disability claims is set for 9 March 2027. The merits have not been decided — there is no finding that Workday discriminated. The small-business takeaway is the one rival guides state well: buying the software does not transfer the liability, and a vendor that cannot explain a rejection leaves the employer carrying the risk.

What it costs if you get it wrong

Penalties are modest on paper and large in aggregate, because most of these laws count each day — or each candidate — separately. Illinois civil penalties run up to $16,000 per act per aggrieved party for a first offence, $42,500 for a second within five years, and $70,000 for two or more within seven years (775 ILCS 5/8A-104(K)), on top of actual damages, attorney’s fees and back pay. New York City’s are up to $500 for a first violation and $500–$1,500 for each subsequent one — but each day a tool is used in violation is a separate violation, and each missed candidate notice is a separate violation too, so a quarter of un-audited screening can reach six figures. Colorado’s are up to $20,000 per violation. Texas’s reach $200,000 per uncurable violation. None of these is a reason to stop hiring. All of them are a reason to add the notice line and name the human reviewer now, while it is still cheap.

What to do before 1 January 2027

  1. List every tool that touches a hiring decision — ATS ranking, résumé filters, video-interview analysis, scheduling filters — and mark which ones score, rank or reject. That list is the whole compliance map.
  2. Add one plain-language notice line to job posts and the application flow: that AI is used, what it does, and how a candidate can ask a human to look.
  3. Name a human reviewer for every rejection the tool influences, and make sure the reviewer actually sees the file before the rejection goes out.
  4. Ask your ATS vendor the five questions above, in writing, and keep the reply. If the vendor cannot answer, that is information too.
  5. Keep records for the strictest place you hire — four years if you hire in California, three in Colorado — and keep the criteria, the tool’s output and who made the call.
  6. Check the two traps: don’t let a tool use ZIP code (or anything like it) as a filter — Illinois bans it expressly — and don’t assume a general-purpose chatbot is covered, or that a ranking tool is not.
  7. Re-check the dates in January. Colorado’s rules are due by 1 January 2027 and its enforcement is paused pending rulemaking; Illinois’s notice rules were proposed in May 2026 and withdrawn in June 2026, so the form of notice is still unsettled even though the duty applies.

The balanced view

These laws are not a ban on AI in hiring, and none of them requires a small business to stop using the tools that make hiring manageable. Illinois’s rule is the strictest in one sense — it makes a discriminatory effect a violation even with no discriminatory intent — but it is also the most familiar: it sits inside a civil-rights law that has governed hiring for decades. Colorado’s rewrite is a genuine softening: the heavy paperwork is gone, and what remains is notice and human review, which most careful employers already do. The real cost for a small firm is not the fine; it is discovering, after a complaint, that nobody can explain why a candidate was rejected. The fixes in this article cost an afternoon. A complaint costs months.

FAQ

Is AI résumé screening legal?

Yes — in every US state, provided the tool does not discriminate and you follow the notice and record rules where they apply. No state bans AI screening outright. Illinois bans AI that has a discriminatory effect, and bans ZIP code as a proxy; Texas bans AI deployed with the intent to discriminate. The question is not whether you may screen with AI, but whether you can explain the result.

Is it illegal to use AI to screen résumés?

No. What is illegal is discrimination in hiring, with or without AI. The new laws add duties around AI specifically — notice, audits, human review — but none of them makes AI screening itself unlawful.

Does Illinois HB 3773 apply to small businesses?

Yes, in almost every case. It covers any employer with one or more Illinois employees for 20 or more calendar weeks in the year of, or the year before, the alleged violation (775 ILCS 5/2-101(B)(1)). There is no separate small-business exemption for the AI provisions.

When does the Colorado AI Act take effect?

The rewritten law, SB 26-189, applies to decisions made on or after 1 January 2027. It was signed 14 May 2026. The original 2024 act never took effect.

Is Colorado’s AI hiring law in effect? Is enforcement paused?

The statute is on the books with a 1 January 2027 start date, but enforcement is paused: a court stayed enforcement of the old act in April 2026, and the attorney general has said he will not enforce the new law until rulemaking is finished. Prepare as if the date stands; do not assume it will be delayed again.

What does NYC Local Law 144 require?

An independent bias audit of the tool within one year before use, a public summary of that audit on your careers page, notice to candidates at least 10 business days before use, and an alternative selection process on request. The duty sits on the employer, not the vendor. It has been enforced since 5 July 2023.

Do California’s ADMT rules apply to my business?

Probably not, if you are small. They apply only to CCPA “businesses” — revenue over $26,625,000, personal information of 100,000+ consumers or households, or 50%+ of revenue from selling or sharing personal information. California’s older FEHA anti-discrimination rules are separate and do apply to employers with five or more employees.

Can my vendor be liable for AI hiring discrimination?

Possibly — that is the live question in Mobley v. Workday, where a federal judge allowed claims against the vendor to proceed on an agency theory. The merits have not been decided, and a class certification hearing is set for 9 March 2027. Do not assume your vendor’s liability protects you: the employer remains responsible for the outcome of its own hiring decisions.

What is a bias audit, and do I need one?

An independent evaluation of whether a hiring tool produces different outcomes for different protected groups, measured by selection rates and impact ratios. Only New York City requires one by name. In other states, documented bias testing is not required but is the best defence if a discrimination claim is made — Connecticut expressly treats it as mitigation.

Do I have to tell candidates AI screened their application?

In Illinois, yes — notice is a statutory duty since 1 January 2026, though the exact form of notice is unsettled because the implementing rules were withdrawn. In Colorado, yes from 1 January 2027. In New York City, yes, at least 10 business days before use. In Connecticut, written notice is required for technology deployed on or after 1 October 2027.

Does the law apply if I only use AI to draft job posts?

Generally no. Drafting, summarising, translating and scheduling are not covered decisions in most jurisdictions — Colorado expressly excludes tools that only do those things for human review. The duties attach when the tool scores, ranks, filters or rejects a person.

What records do I have to keep, and for how long?

Keep the criteria, the tool’s output and who made the call. California’s FEHA rules require at least four years; Colorado requires three years. Keep the strictest period that applies anywhere you hire.

Sources

  • Illinois General Assembly — Public Act 103-0804 (HB 3773); 775 ILCS 5/2-101(B)(1) (employer definition) and 5/8A-104(K) (civil penalties)
  • Colorado General Assembly — SB 26-189 fiscal note and enrolled text (leg.colorado.gov); signed 14 May 2026
  • Texas Legislature — HB 149 (TRAIGA), enrolled text, Tex. Bus. & Com. Code §§ 552.104–552.105 (penalties)
  • NYC Department of Consumer and Worker Protection — Local Law 144 FAQ and Admin Code §§ 20-870 to 20-874 (§ 20-872 penalties)
  • Regulation (EU) 2024/1689 — EU AI Act, high-risk hiring duties from 2 December 2027 (see our EU AI Act guide)
  • Mobley v. Workday, No. 3:23-cv-00770-RFL (N.D. Cal.) — 12 July 2024 order; 16 May 2025 collective certification
  • Illinois Department of Human Rights — proposed notice rules, 15 May 2026; withdrawn 2 June 2026 (no revised proposal as of 2 Nov 2026)

This article is general information, not legal advice. Laws change — Colorado’s already moved twice — and how they apply depends on where your candidates live and what your tools actually do. A lawyer should check any system that touches hiring in the places you hire. Figures checked 15 November 2026 unless noted.

Related guides: AI for Small Business: Implementation Guide · Client Intake Automation Checklist for Freelancers · Best AI Tools for Small Business 2026

Was this article helpful?

Leave a Comment