# Using AI in hiring responsibly, a checklist

> Using AI in hiring responsibly means telling candidates where AI is used, keeping a person on every decision, checking outcomes for adverse impact, offering adjustments, and keeping only the data you need. Rules to know early include New York City's Local Law 144, Illinois's Artificial Intelligence Video Interview Act and the EU AI Act, which lists recruitment as a high-risk use; this checklist is not legal advice.

Published by Emoti Research on 2026-11-01, updated 2026-11-01. Article. For talent leaders. Canonical: https://emoti.ai/articles/using-ai-in-hiring-responsibly

## What does responsible use of AI in hiring involve?

Five habits, kept every time a tool touches a candidate: tell people, keep a person on every decision, check outcomes for adverse impact, offer adjustments, and collect and keep only the data you need.

Use this checklist before you switch a tool on, and again every few months:

1. Candidates are told, in plain words and before the AI step, what the tool does and who reviews its output.
2. A named person makes every advance or reject decision and can see the evidence behind it.
3. Selection rates are compared by group at each stage the tool affects.
4. Candidates know how to ask for an adjustment or an alternative, and requests are answered quickly.
5. You know what data the tool collects, why, where it goes and when it is deleted.
6. You know which laws apply where you hire, and who is responsible for each duty.

This article is not legal advice. Rules change, and the official pages linked below are the place to check the current position.

## How should you tell candidates?

Before the AI step, say what the tool does, what it evaluates, who sees the result, and how to ask for an alternative or an adjustment.

Plain words work better than legal ones: "A recorded AI interviewer will ask you five questions about your work. People on our recruiting team review every answer before any decision." Some laws set minimums. New York City's Local Law 144 requires employers and employment agencies to tell candidates who live in the city, 10 business days before using an automated employment decision tool, that the tool will be used and which job qualifications and characteristics it will assess, with instructions for requesting a reasonable accommodation. Illinois's Artificial Intelligence Video Interview Act requires employers that use AI to analyze recorded video interviews for Illinois-based positions to notify applicants, explain how the AI works and what general types of characteristics it uses, and obtain consent before the interview.

## How do you keep a person on every decision?

Make sure a named person decides each advance or reject, sees the evidence the tool used, and has the time and authority to disagree with it. A tool should never reject anyone on its own.

Human review only counts if it is real. A recruiter who approves 200 recommendations in ten minutes is not reviewing them. Give reviewers the underlying evidence as well as any score; track how often they disagree with the tool; and treat a disagreement rate near zero as a warning rather than a success.

In the EU, the GDPR gives people the right not to be subject to a decision based solely on automated processing that produces legal effects for them or similarly significantly affects them. Where such a decision is allowed because it is necessary for a contract or the person has given explicit consent, they must still be able to obtain human intervention, give their point of view and contest the decision (Article 22).

## How do you check for adverse impact?

Compare the selection rate of each group at each stage. Under the US Uniform Guidelines on Employee Selection Procedures, a selection rate for any race, sex or ethnic group that is less than four-fifths (80 percent) of the rate for the group with the highest rate is generally regarded as evidence of adverse impact (29 CFR 1607.4(D)).

An illustrative calculation, with made-up numbers:

| Group | Applied | Advanced | Selection rate | Ratio to highest rate |
| --- | --- | --- | --- | --- |
| Group A | 100 | 30 | 30% | 1.00 |
| Group B | 90 | 18 | 20% | 0.67 |

Group B's ratio of 0.67 is below 0.80, so the stage would be flagged for a closer look. The guideline is a rule of thumb, not a verdict. The same section says smaller differences can still be adverse impact when they are significant in both statistical and practical terms, and larger ones may not be when they rest on small numbers.

New York City's law turns this into a duty for covered tools: an independent bias audit within one year before use, with selection or scoring rates and impact ratios by sex, by race and ethnicity, and by intersectional categories, and a published summary of the results.

## How should you handle adjustments and data?

Tell candidates how to ask for an adjustment and answer quickly, and collect only the data a decision needs, for only as long as you need it.

### Adjustments

An AI screening step can quietly assume a candidate who hears well, speaks fluently and is comfortable being timed. That is not everyone. In the US, the Americans with Disabilities Act counts not making reasonable accommodations to the known limitations of a qualified applicant with a disability as discrimination, unless the accommodation would impose an undue hardship. In Great Britain, the Equality Act 2010 places a duty on employers to make reasonable adjustments. Offer a real alternative, such as a live interview, written answers or extra time, and put a sentence in every invitation: "If you'd like an adjustment or a different format, reply to this email."

### Data

For each tool, write down what it collects (audio, video, transcripts, scores), why, who can see it, and when it is deleted. The GDPR requires personal data to be limited to what is necessary for its purpose and kept no longer than necessary (Article 5). Illinois's video interview law limits sharing applicant videos to people whose expertise or technology is needed to evaluate the applicant, and requires employers to delete an applicant's interviews within 30 days of the applicant's request.

## Which rules do teams meet first?

Three are a good place to start: New York City's Local Law 144, the Illinois Artificial Intelligence Video Interview Act and the EU AI Act. Other places have their own rules, so check where you hire.

| Rule | Where it applies | What it requires | Status |
| --- | --- | --- | --- |
| Local Law 144 of 2021 | Automated employment decision tools used to assess candidates for jobs in New York City | An independent bias audit within a year before use, a published summary, and notice to candidates 10 business days before use | Enforced since July 5, 2023 |
| Artificial Intelligence Video Interview Act (820 ILCS 42) | AI analysis of recorded video interviews for positions based in Illinois | Notice, an explanation of how the AI works, consent, limits on sharing and deletion on request; demographic reporting for employers that rely solely on AI to choose who gets an in-person interview | In effect since January 1, 2020; reporting since January 1, 2022 |
| EU AI Act (Regulation (EU) 2024/1689) | AI systems used in the EU, including for recruitment and selection | Recruitment and selection systems are listed as high-risk in Annex III; AI that infers emotions in the workplace is prohibited | High-risk obligations for Annex III systems apply from December 2, 2027, under amending Regulation (EU) 2026/1744; the prohibition has applied since February 2, 2025 |

A few details matter. New York City's Department of Consumer and Worker Protection says the law's requirements apply when a tool is used to assess people who have applied for a specific position, not when it is used to scan a resume bank, reach out to potential candidates or invite applications. The vendor of a tool is not responsible for its bias audit under that law; the employer is. And the European Commission's guidelines on prohibited practices treat hiring as part of the workplace for the emotion recognition ban.

### Where Emoti fits

Emoti is designed around the same habits, and it doesn't do the employer's legal work. Every ranking shows its evidence, criterion by criterion. Emoti never advances or rejects anyone; a person makes each decision. Voice screens tell candidates that an AI interviewer asks the questions and that people review every answer, and they rate the content of answers against the team's rubric, not the candidate's face or tone of voice. Our research papers describe each method's limits, including gaps we haven't closed yet, such as the lack of a separate consent step in voice screens: see [Evidence-first candidate matching](https://emoti.ai/research/evidence-first-matching) and [Designing first-round voice screens candidates can trust](https://emoti.ai/research/trustworthy-voice-screens). Notices, audits and decisions about adjustments remain the employer's responsibility.

## Key terms

- **Adverse impact**: A substantially lower selection rate for one group, defined by race, sex or ethnicity, compared with another at a step in the hiring process.
- **Four-fifths rule**: A guideline in the US Uniform Guidelines on Employee Selection Procedures: a selection rate for any race, sex or ethnic group below four-fifths of the highest group's rate is generally regarded as evidence of adverse impact.
- **Automated employment decision tool**: A term from New York City's Local Law 144 for a computer-based tool that uses machine learning, statistical modeling, data analytics or artificial intelligence to substantially assist or replace discretionary hiring or promotion decisions.
- **Bias audit**: An impartial evaluation of a hiring tool's outcomes by group. Under New York City's Local Law 144, an independent auditor calculates selection or scoring rates and impact ratios by sex, race and ethnicity, and their intersections.
- **Reasonable adjustment**: A change to a hiring process or job that removes a disadvantage for a disabled person, such as extra time, a different format or a human interviewer. US law calls it a reasonable accommodation.

## Questions

### Does keeping a person in the loop remove legal risk?

No. Anti-discrimination law applies to decisions people make with a tool's help, and a reviewer who rubber-stamps a tool's output adds little protection. Review helps when the person sees the evidence and can disagree.

### How often should we check AI hiring tools for adverse impact?

Whenever a tool or its settings change, and on a regular schedule, such as every quarter, at each stage the tool affects. With small numbers, look at longer periods rather than single months.

### Can a bias audit use inferred demographic data?

Not under New York City's rules. Its FAQ says imputed or inferred demographic data cannot be used to conduct a bias audit, and that test data can be used when there isn't enough historical data.

### What should we ask an AI hiring vendor?

What the tool evaluates and how, what data it uses and keeps, how a person reviews its output, what bias testing has been done and on what data, and how candidates can ask for an alternative.

## Sources

- US Equal Employment Opportunity Commission (1978). Uniform Guidelines on Employee Selection Procedures, 29 CFR 1607.4, Information on impact. Electronic Code of Federal Regulations. https://www.ecfr.gov/current/title-29/subtitle-B/chapter-XIV/part-1607/section-1607.4
- NYC Department of Consumer and Worker Protection (2023). Automated Employment Decision Tools (AEDT). City of New York. https://www.nyc.gov/site/dca/about/automated-employment-decision-tools.page
- NYC Department of Consumer and Worker Protection (2023). Automated Employment Decision Tools: Frequently Asked Questions. City of New York. https://www.nyc.gov/assets/dca/downloads/pdf/about/DCWP-AEDT-FAQ.pdf
- Illinois General Assembly (2019). Artificial Intelligence Video Interview Act (820 ILCS 42). Illinois Compiled Statutes. https://www.ilga.gov/legislation/ilcs/ilcs3.asp?ActID=4015&ChapterID=68
- European Parliament and Council of the European Union (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- European Parliament and Council of the European Union (2026). Regulation (EU) 2026/1744 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2026/1744/oj
- European Commission (2025). Commission publishes the Guidelines on prohibited artificial intelligence (AI) practices, as defined by the AI Act. European Commission. https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-prohibited-artificial-intelligence-ai-practices-defined-ai-act
- Dushi, D. (2026). Red Lines under EU AI Act: Unpacking the prohibition of emotion recognition in the workplace and education institutions. Future of Privacy Forum. https://fpf.org/?p=248772
- European Parliament and Council of the European Union (2016). Regulation (EU) 2016/679 (General Data Protection Regulation), Articles 5 and 22. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2016/679/oj
- United States Congress (1990). Americans with Disabilities Act, 42 U.S.C. § 12112. United States Code, via govinfo.gov. https://www.govinfo.gov/content/pkg/USCODE-2010-title42/pdf/USCODE-2010-title42-chap126-subchapI-sec12112.pdf
- UK Parliament (2010). Equality Act 2010, section 20: Duty to make adjustments. legislation.gov.uk. https://www.legislation.gov.uk/ukpga/2010/15/section/20