Consent and compliance

AI hiring laws for staffing agencies (2026): employment agency, deployer or developer, and who gives the notice

On this page
  1. Which AI hiring laws apply to staffing and recruiting agencies?
  2. Is your agency the employer, the employment agency, the deployer or the developer?
  3. Who gives the candidate notice when an agency screens for a client?
  4. Who is responsible for the bias audit or bias testing?
  5. Can an agency apply a client's AI screening criteria?
  6. How is liability shared between agency, client and vendor?
  7. What records should an agency keep?
  8. What changes when the agency recruits in the UK or the EU?
  9. A copy-ready AI responsibilities schedule for client contracts
  10. Checklist for staffing and recruiting agencies
  11. Questions people ask

AI hiring laws reach staffing and recruiting agencies directly, not only through their clients. New York City's Local Law 144 applies to "an employer or an employment agency" that uses an automated employment decision tool; California's civil rights regulations, in force since October 1, 2025, treat anyone running recruitment or screening for an employer through an automated-decision system as that employer's agent, and therefore an employer, and define employment agencies to include procurement of applicants through such a system; and Colorado (from January 1, 2027) and Connecticut (for tools deployed from October 1, 2027) put their notice duties on the "deployer," which is the agency whenever the agency runs the tool. Under the EU AI Act an agency using high-risk recruitment AI is a deployer, and becomes the provider if it puts its own name on the system. All of this sits on top of the federal rule that an employment agency may not refer or classify people on a protected basis (42 U.S.C. § 2000e-2(b)), as of November 2026.

Not legal advice. This page summarizes the statutes, regulations and regulator pages linked in each row, as checked in November 2026. Several of these laws are new, the Colorado and Connecticut duties have not started, and none of them was written with every staffing model in mind. Check your own contracts and states with counsel.

Which AI hiring laws apply to staffing and recruiting agencies?

RuleWho it coversWhat it requires of an agencyIn force fromSource
Title VII "Employment agency": anyone regularly undertaking, with or without compensation, to procure employees for an employer or work for employees, including its agents No refusing to refer, classifying or referring on the basis of race, color, religion, sex or national origin, whether a person or a tool does it In force 42 U.S.C. 2000e(c); 2000e-2(b)
ADEA recordkeeping Employment agencies Keep placements, referrals, job orders, applications and agency-administered test papers for one year In force 29 CFR 1627.4
New York City Local Law 144 Employers and employment agencies using an AEDT "in the city" Bias audit within one year before use, public summary, notice to New York City resident candidates at least 10 business days before use Enforced since July 5, 2023 Admin. Code 20-871; DCWP FAQ
California FEHA regulations Agents acting for employers, "including when such activities and decisions are conducted in whole or in part through the use of an automated decision system"; employment agencies, including those working "through the use of an automated-decision system" No discriminatory use of an automated-decision system; anti-bias testing counts as evidence; keep automated-decision system data four years October 1, 2025 2 CCR 11008(a), (g), 11009(f), 11013(c)
Illinois Human Rights Act Employers (2-102(L)); employment agencies (2-102(B)) Employers: no AI with a discriminatory effect, no zip codes as a proxy, notice of AI use. Agencies: no discriminatory referral and no accepting job orders that have the effect of unlawful discrimination AI provision January 1, 2026 775 ILCS 5/2-101, 2-102
Illinois AI Video Interview Act "An employer" that asks applicants to record video interviews and uses AI to analyze them, for Illinois-based positions Notice, an explanation of how the AI works, consent, sharing limits, deletion within 30 days of a request January 1, 2020 820 ILCS 42
Colorado SB 26-189 "Deployer": a person doing business in Colorado that deploys a covered ADMT; "developer" includes anyone who intentionally and substantially modifies an ADMT so it becomes covered Notice before use, disclosure within 30 days after an adverse outcome, correction and human review, three years of records Consequential decisions on or after January 1, 2027 C.R.S. 6-1-1701 to 1709
Connecticut Public Act 26-15 "Deployer": a person doing business in the state who deploys an automated employment-related decision technology in the state Tell people when they interact with the technology; a written pre-decision notice with six listed items Deployments on or after October 1, 2027 P.A. 26-15, secs. 7 to 12
Texas Responsible AI Governance Act "Deployer": a person who deploys an AI system for use in Texas No developing or deploying AI "with the intent to unlawfully discriminate"; answer attorney general demands for documentation January 1, 2026 Bus. & Com. Code 552.056, 552.103
EU AI Act Deployers of high-risk AI, which includes recruitment and candidate evaluation (Annex III, point 4(a)); a deployer becomes a provider under Article 25 Follow instructions for use, human oversight, monitoring, six months of logs, inform workers and candidates 2 December 2027 for Annex III systems Art. 26; Art. 25
UK GDPR (as amended) Controllers taking significant decisions based solely on automated processing Safeguards: information, representations, human intervention, contest; clear controller and processor roles 5 February 2026 DUAA s. 80

Is your agency the employer, the employment agency, the deployer or the developer?

Each law uses its own label, and one agency can carry several at once. Work out each role separately:

  • Employment agency. Title VII's definition covers any person "regularly undertaking with or without compensation to procure employees for an employer or to procure for employees opportunities to work for an employer." Illinois' covers anyone regularly undertaking "to procure opportunities to work or to procure, recruit, refer, or place employees." Direct-hire recruiters, staffing firms and RPO providers all fit.
  • Employer. A staffing firm is the employer of the temps on its payroll, so employer duties, such as Illinois' notice of AI use under 2-102(L), can apply to decisions about them. In California an agent doing screening for an employer through an automated-decision system "is also an 'employer.'" Joint employment with the client is covered in joint employer liability for staffing agencies.
  • Deployer. In Colorado, Connecticut, Texas and the EU, the deployer is whoever puts the tool to use. If your recruiters run the screening tool, you are the deployer, whatever the client contract says.
  • Developer or provider. An agency that builds its own matching or scoring model, or substantially modifies a vendor's, can become a developer in Colorado and Connecticut. Under the EU AI Act a deployer becomes the provider if it puts "their name or trademark" on a high-risk system, makes a substantial modification, or changes a system's intended purpose so it becomes high-risk (Article 25(1)). White-labelling a vendor's assessment for clients is the common trap.

Whether a tool is covered at all depends on what it outputs and how much weight the output carries; the definitions are compared in what an automated employment decision tool is.

Who gives the candidate notice when an agency screens for a client?

The party that uses the tool. In practice that splits three ways:

  1. The agency runs the tool (resume ranking, an assessment, a chatbot screen) before submitting candidates. The agency gives the notices. In New York City, DCWP's FAQ says an agency's use is "in the city" when the agency is located in New York City or, for agencies elsewhere, when the job is in a New York City office or a remote job associated with one; notice goes to candidates who live in the city at least 10 business days before use. In Colorado, the agency as deployer gives the point-of-interaction notice and, within 30 days after an adverse outcome, the disclosure in C.R.S. 6-1-1704(3).
  2. The client runs its own tool on candidates the agency submits. The client is the user and gives its notices. The agency should still tell candidates where their application goes, under its own privacy notice.
  3. Both use tools at different stages. Each gives notice for its own stage. Agree the wording so candidates are not told contradictory things.

Connecticut adds a contract option: a developer may agree with a deployer to take over the deployer's notice duties, and the contract must say which ones (P.A. 26-15, sec. 8(c)). The Illinois AI Video Interview Act is written for "an employer"; an agency that asks applicants to record AI-analyzed video interviews for Illinois-based positions should give the notice, explanation and consent steps rather than rely on that wording. Templates are in AI disclosure to candidates, and the wider question is covered in do you have to tell candidates you use AI.

Who is responsible for the bias audit or bias testing?

In New York City, the user. DCWP's FAQ says employers and employment agencies are responsible for making sure a bias audit was done before use, the vendor is not, and a vendor may commission an independent audit of its tool. An agency can rely on an audit built from several users' data only if it supplied its own historical data to the auditor or it is using the tool for the first time. Elsewhere testing is evidence rather than a duty: California's regulation makes anti-bias testing, its quality, recency and scope, and the response to the results relevant to a claim or defense (2 CCR 11009(f)), and Connecticut lets the commission or a court consider the same evidence. The audit itself is explained in the Local Law 144 bias audit guide.

Can an agency apply a client's AI screening criteria?

Not if the criteria discriminate, and a client's instruction is no defense. Title VII makes it unlawful for an employment agency "to classify or refer for employment any individual on the basis of" race, color, religion, sex or national origin. The Illinois Human Rights Act bars an agency from accepting "any job order, requisition or request for referral of applicants" that "makes or has the effect of making" unlawful discrimination, and its AI provision names zip codes as a forbidden proxy. California defines a proxy as "a characteristic or category closely correlated with a basis protected by the Act" (2 CCR 11008(l)). Before you load a client's knockout rules or score cutoffs into a tool, check them for proxies such as graduation dates, zip codes and employment gaps; the method is in Title VII disparate impact and AI tools.

How is liability shared between agency, client and vendor?

  • California: the agent running screening through an automated-decision system is itself an employer under FEHA, so the agency cannot treat liability as the client's alone.
  • Colorado: in discrimination actions arising from a decision materially influenced by covered ADMT, fault is allocated among deployers and developers by relative fault, and, subject to the exceptions in the section, a contract clause that indemnifies a developer or deployer against damages for its own acts or omissions in violation of Colorado anti-discrimination law is void (C.R.S. 6-1-1707). Read your vendor agreements before renewal.
  • Federal: Title VII's definition of employment agency includes "an agent of such a person," and the client remains an employer for its own decisions.

What records should an agency keep?

Keep the tool's outputs with version and date, the human decision, the notices given, candidate requests, and the audit or test evidence, for the longest period any applicable law sets: four years where California applies, three years after each decision under Colorado's law from 2027, and the federal one-year minimum for agency records under 29 CFR 1627.4. The full table is in AI hiring recordkeeping requirements.

What changes when the agency recruits in the UK or the EU?

In the UK, the questions are data protection ones: is the agency a controller or a processor for each use of the tool, who tells the candidate, and whether a decision is solely automated, which triggers the Article 22C safeguards. The ICO found AI providers wrongly calling themselves processors and contracts that left neither side responsible for telling candidates; see ICO guidance on AI in recruitment. In the EU, an agency using recruitment AI is a deployer with Article 26 duties from 2 December 2027 (EU AI Act guide).

A copy-ready AI responsibilities schedule for client contracts

AI TOOL RESPONSIBILITIES SCHEDULE (example: review with counsel before use)

Tool and feature:            [vendor, product, feature, version]
Who runs it:                 [Agency / Client / both at stages: describe]
Roles under each law:        NYC LL144 user: [Agency/Client]
                             Colorado / Connecticut deployer: [Agency/Client]
                             EU AI Act deployer or provider: [Agency/Client]
Jurisdictions in play:       [job locations; candidate locations]
Candidate notices given by:  [party], wording approved by [party], sent [when]
Bias audit / testing:        commissioned by [party]; historical data supplied
                             by [party]; summary posted at [URL]
Screening criteria:          supplied by [Client]; reviewed for proxies by
                             [Agency] on [date]; changes need written approval
Human review:                [named role] reviews every candidate at [stage]
Adverse-outcome disclosures: sent by [party] within [30] days (Colorado)
Records kept:                [what], by [party], for [period]
Candidate requests:          opt-out, accommodation, data, deletion handled by
                             [party] within [time]; forwarded within [time]
Vendor documentation:        held by [party]; shared on request
Review:                      [date]; on any tool or criteria change

Checklist for staffing and recruiting agencies

  1. List every tool your recruiters use that scores, ranks, filters or recommends candidates, and every tool your clients run on candidates you submit.
  2. For each, record your role under each law in the table: employment agency, employer, deployer, developer or provider.
  3. Check where your offices, jobs and candidates are: New York City, California, Illinois, Colorado, Connecticut, Texas, the UK, the EU.
  4. For New York City use, confirm a bias audit from the past year, the public summary and the 10-business-day notice.
  5. Review client screening criteria for proxies before loading them into any tool.
  6. Name a human reviewer who sees every candidate at each automated stage, and record the review.
  7. Sign a responsibilities schedule with each client and each vendor; remove indemnities Colorado makes void.
  8. Set retention to the longest applicable period and align vendor auto-delete settings.
  9. Prepare Colorado notices and adverse-outcome disclosures before January 1, 2027, and Connecticut notices before October 1, 2027.
  10. Re-check this list when the Colorado attorney general's rules and the final EU and UK guidance are published.

The state-by-state picture is in AI hiring laws by state, with deep guides to California, Colorado, Illinois' AI amendment and Texas. For governance inside the agency, start from the AI governance policy for recruiting.

Questions people ask

Does NYC Local Law 144 apply to staffing agencies?

Yes. The law applies to "an employer or an employment agency" that uses an automated employment decision tool to screen candidates in the city. DCWP's FAQ says an agency's use is "in the city" if the agency is located in New York City, or, if it is outside the city, if the job is in a New York City office or is a remote job associated with one. The agency must make sure a bias audit was done in the past year, publish the summary, and give notice to candidates who live in the city.

If our client's AI tool screens the candidates we submit, who is responsible?

Usually the party that uses the tool to screen. Under New York City's law that is the employer or agency using the tool; under Colorado's and Connecticut's laws it is the deployer, the person doing business in the state that puts the tool into use. An agency that only submits candidates to a client's own tool is not the user of that tool, but it remains liable for its own referrals, so agree in writing who gives which notice.

Is a staffing agency a developer if it builds its own candidate-matching model?

It can be. Colorado's SB 26-189 makes a developer anyone doing business in Colorado who develops or makes available a covered ADMT, or who intentionally and substantially modifies one so it becomes covered; Connecticut's Public Act 26-15 and the EU AI Act have similar rules, and under the EU AI Act an agency that puts its own name on a high-risk system becomes its provider.

Can a staffing agency apply a client's AI screening criteria?

Only if the criteria are lawful. Title VII makes it unlawful for an employment agency to refer or classify people on the basis of race, color, religion, sex or national origin, the Illinois Human Rights Act bars agencies from accepting job orders that have the effect of unlawful discrimination, and Illinois expressly bars using zip codes as a proxy for protected classes in AI used for hiring.

Do AI hiring laws require a staffing agency to keep records?

Several do. California requires employers and covered entities, including employment agencies, to keep automated-decision system data for four years; federal ADEA rules require employment agencies to keep placement, referral, job order and application records for one year; and Colorado will require deployers to keep compliance records for three years after each decision from January 1, 2027.