AI Hiring Tools: Minnesota Employer Compliance

What makes AI candidate screening lawful in Minnesota?

AI candidate screening is lawful in Minnesota because the state has not enacted a law that bans or specifically regulates artificial intelligence in hiring, so using AI to screen or rank candidates is allowed in principle. The catch is that the tool inherits every legal duty a human recruiter carries. The MHRA makes it an unfair employment practice to “refuse to hire or to maintain a system of employment which unreasonably excludes a person seeking employment” on a protected basis (Minn. Stat. § 363A.08, subd. 2). A screening algorithm is a system of employment. If it filters out protected-class applicants without a sound job-related reason, the employer has the problem, not the software company. Treat an AI hiring tool with the same legal care as any other employment decision: the question is never “is AI allowed,” but “can I justify the specific decisions this tool produced.” In my experience, the employers who get into trouble bought a tool for speed and never asked the vendor how it was tested for bias.

How does the EEOC’s 2025 AI guidance rollback affect your obligations?

It does not reduce them. The Equal Employment Opportunity Commission (“EEOC”) removed its 2022 and 2023 technical-assistance documents on AI in hiring from its website, but taking down guidance does not repeal a statute. Those documents explained how the agency read existing law; they were never the law themselves. Title VII of the Civil Rights Act still bans employment practices that cause a disparate impact, written into the statute at 42 U.S.C. § 2000e-2, and the MHRA, the Americans with Disabilities Act (“ADA”), and the Fair Credit Reporting Act (“FCRA”) all still apply to algorithmic hiring. The practical effect of the rollback is less federal instruction, not less liability. Several jurisdictions have moved the other direction: New York City requires bias audits of automated employment decision tools, Illinois has added AI-notice duties to its Human Rights Act, and Colorado has enacted its own AI law, which it has since revised. Minnesota has done none of this, so Minnesota employers answer to the general antidiscrimination statutes rather than a dedicated AI rulebook. For a sense of how quickly this area moves, see the recent changes to Minnesota employment law.

How can an AI hiring tool create disparate-impact liability?

An AI hiring tool creates disparate-impact liability when it selects one protected group at a meaningfully lower rate than another and the employer cannot justify the practice. Under the disparate-impact provision of Title VII, 42 U.S.C. § 2000e-2(k), a practice that causes a disparate impact is unlawful unless the employer shows it is “job related for the position in question and consistent with business necessity.” Disparate impact does not require intent. A tool can be built with no protected characteristic as an input and still produce an unlawful result, because it learns from data that reflects who was hired before. Federal agencies measure this with the four-fifths rule: under the Uniform Guidelines on Employee Selection Procedures (“Uniform Guidelines”), a selection rate for any group “which is less than four-fifths (4/5) (or eighty percent) of the rate for the group with the highest rate will generally be regarded . . . as evidence of adverse impact” (29 C.F.R. § 1607.4(D)). In plain terms: if your tool advances 80 women for every 100 men it advances, and you cannot show the screen actually predicts job performance, you are exposed. Ask any vendor for that group-by-group selection data before you deploy, then re-run it on your own applicant pool.

How does the Minnesota Human Rights Act apply to AI screening tools?

The MHRA applies to an AI screening tool the same way it applies to any hiring method, and it covers more employers and more protected classes than federal law. It is an unfair employment practice to “maintain a system of employment which unreasonably excludes a person seeking employment” on a protected basis (Minn. Stat. § 363A.08, subd. 2). Two features make this the statute Minnesota employers should watch most closely. First, coverage: the MHRA defines an “employer” as “a person who has one or more employees” (Minn. Stat. § 363A.03, subd. 16), so a Minnesota company too small for Title VII or the ADA, which start at 15 employees, is still covered by the Act’s general antidiscrimination prohibition. (Its reasonable-accommodation duty, taken up below, is the exception: like the ADA, that duty starts at 15 or more employees.) Second, protected classes: the MHRA lists more of them than Title VII, adding creed, marital status, status with regard to public assistance, sexual orientation, gender identity, and familial status. A tool that screens on any of those, or on a proxy for them, reaches Minnesota law even where federal law might not. The same care you would give to accommodating religious dress and practice applies to how an algorithm treats every protected group on that longer Minnesota list.

How do AI tools screen out applicants with disabilities?

AI assessments screen out applicants with disabilities when they measure a trait tied to a disability rather than to the job, and that is the fastest-growing risk in AI hiring. The ADA prohibits selection criteria “that screen out or tend to screen out an individual with a disability” unless the criteria are job-related and consistent with business necessity, and it also requires an employer to administer tests so results reflect the skill measured rather than an applicant’s impaired sensory, manual, or speaking ability (42 U.S.C. § 12112, subds. (b)(6) and (b)(7)). A video-interview tool that scores eye contact or speech patterns can penalize an autistic applicant or someone with a speech disability; a timed game can penalize someone with a motor or cognitive impairment. Minnesota law adds its own accommodation duty, though only for larger employers. For a business with 15 or more employees (the same threshold the ADA uses), it is an unfair employment practice “not to provide a reasonable accommodation for a job applicant . . . with a disability” unless doing so would impose an undue hardship (Minn. Stat. § 363A.08, subd. 6). The practical rule I give clients: every AI assessment needs a visible, simple way for an applicant to request a human alternative, and you should know your Minnesota disability discrimination rules before you turn one on.

When does the federal FCRA apply to AI candidate scoring?

The FCRA applies when your AI score comes from a third party that assembles information about the applicant, not when your own software scores your own application data. If a background-check or scoring vendor qualifies as a consumer reporting agency, its output is a “consumer report,” and the FCRA’s hiring rules apply. Those rules are specific. Before you obtain the report, you must give the applicant “a clear and conspicuous disclosure . . . in writing . . . in a document that consists solely of the disclosure,” and the applicant must have “authorized in writing” the report (15 U.S.C. § 1681b, (b)(2)(A)). The standalone-document requirement trips up employers who bury the disclosure inside the application form. Then, before you reject anyone based on the report, you must provide “a copy of the report” and “a description in writing of the rights of the consumer” (§ 1681b(b)(3)(A)), so the applicant can dispute an error first. The line that matters for AI: a tool that only analyzes resumes you already hold is usually not a consumer report, but a vendor that pulls outside data, such as social media, public records, or cross-employer scoring, often is. When in doubt, treat the vendor as a consumer reporting agency and follow the steps.

How does Minnesota’s consumer data privacy law apply to hiring AI?

It largely does not. The Minnesota Consumer Data Privacy Act (“MCDPA”), effective in 2025, gives privacy rights to a “consumer,” and it defines a consumer as someone acting in a personal capacity, not an employment one. This matters because the MCDPA created a right to opt out of certain automated profiling, and some commentators have suggested that right reaches hiring. The statute’s own definition closes that door: a “consumer” is “a natural person who is a Minnesota resident acting only in an individual or household context,” and “does not include a natural person acting in a commercial or employment context” (Minn. Stat. § 325M.11). An applicant dealing with a prospective employer is acting in an employment context, so the MCDPA’s profiling opt-out does not give that applicant rights against your hiring tool. The takeaway is a warning, not a permission slip: do not treat the MCDPA as your AI hiring compliance plan. Your real obligations run through the antidiscrimination and consumer-reporting statutes above, which the MCDPA does not displace.

What protections belong in your AI hiring vendor contract?

Because the employer keeps the legal liability even when the vendor’s tool causes the harm, your contract for an AI hiring tool should shift as much of that risk back to the vendor as you can. Five terms matter most:

1. Audit rights. The right to inspect, or to have a third party audit, the tool for adverse impact, and to see the results.
2. Validation and adverse-impact data. The vendor’s documentation showing the tool was validated for the job and tested for disparate impact across protected groups, refreshed on a set schedule.
3. Compliance representations and warranties. A written promise that the tool complies with Title VII, the ADA, and the MHRA, plus a duty to notify you of any change that affects compliance.
4. Indemnification. The vendor covers your defense and liability for a discrimination claim traced to the tool’s design.
5. Cooperation and data access. The vendor must preserve and hand over the inputs, scores, and logs you would need to defend a claim.

One scoping point specific to Minnesota: a contract for an AI hiring tool is generally analyzed under Minnesota common-law contract principles rather than the sale-of-goods rules of the Uniform Commercial Code, though a software contract’s classification can turn on the agreement’s predominant purpose. Where common-law principles govern, the enforceability of a liability cap or a warranty disclaimer turns on ordinary contract law. In my experience, vendors resist the indemnity and the audit rights hardest, which tells you exactly which two terms carry the most value.

How do you build a defensible AI hiring process in Minnesota?

A defensible AI hiring process keeps a human accountable for every rejection and creates a record showing each decision was job-related and applied consistently. The goal is not a perfect tool; it is a process you can explain to a judge, an applicant, or the Minnesota Department of Human Rights. Six steps carry most of the weight:

1. Keep a human in the loop. A person, not the software, should own each rejection and hold authority to override the tool. Route close calls to human review by your HR team.
2. Validate before you deploy. Confirm the tool actually predicts performance for the specific job, and get the vendor’s evidence in writing.
3. Test for adverse impact, then re-test. Run the four-fifths measure on your own applicant pool, not just the vendor’s benchmark, and repeat it as your applicants change.
4. Build in an accommodation path. Give every applicant a visible, simple way to request a human alternative to an automated assessment.
5. Apply the tool consistently. Use it the same way for every applicant to the same role; selective use is itself evidence of bias.
6. Document the basis for decisions. Keep the inputs, scores, and the reviewer’s reasons, the same discipline you bring to pay decisions under Minnesota’s pay transparency requirements.

In my practice, the employers who can produce this record rarely see a claim go far, and the ones who cannot are the ones who settle.

Artificial intelligence in hiring is not a new body of law; it is old law pointed at new tools. Minnesota has not passed an AI hiring statute, so the same duties that govern a human recruiter, the Minnesota Human Rights Act, Title VII, the ADA, and the FCRA, govern the algorithm, and the liability stays with you no matter who built the tool. The employers who use these tools well treat them as one input into a human decision they can document and defend, not as the decision-maker. The rules that surround these hiring choices are part of our Minnesota employment law practice. If you are choosing an AI hiring tool, or you have one running and want a read on your exposure, email [email protected] with a short description of the role. I begin every new matter with an intake and conflict check before you send any confidential vendor materials.