Automated employment decision tools (AEDTs) are software systems, often using AI or machine learning, that score, rank, classify or filter candidates in a way that substantially assists or replaces human judgment in hiring or promotion. Examples include resume screening that ranks applicants, video interview scoring and assessment tools that recommend who to advance. Several laws now regulate these tools, most notably New York City's Local Law 144, and general anti-discrimination law applies to them everywhere. This guide explains what counts as an AEDT, the main rules employers should know and a practical compliance checklist. It is general information, not legal advice.
Key takeaways
- An automated employment decision tool (AEDT) is software that scores, ranks or filters candidates in a way that substantially shapes a hiring decision.
- New York City's Local Law 144 requires a yearly independent bias audit, a public summary of the results and advance notice to candidates.
- Federal anti-discrimination law applies to AI-assisted decisions, and employers stay responsible even when a vendor built the tool.
- Keep a person accountable for decisions, document how the tool works and check outcomes for bias regularly.
What counts as an automated employment decision tool?
Under New York City's definition, an AEDT is a computational process, derived from machine learning, statistical modeling, data analytics or artificial intelligence, that issues a simplified output such as a score, classification or recommendation, and that is used to substantially assist or replace discretionary decision making for employment decisions. Other laws use different wording, but the idea is similar: a tool that meaningfully shapes who moves forward.
| Likely in scope | Usually out of scope |
|---|---|
| AI resume screening that ranks or rejects applicants | A job board that only distributes your ad |
| AI scoring of recorded video interview answers used to shortlist | A spreadsheet where a person sorts candidates manually |
| Assessment platforms that recommend who to advance | Scheduling tools and calendar software |
| Chatbots that screen out candidates automatically | Spam filters and basic data storage |
Whether a specific tool is covered depends on how you use it. A score that a recruiter glances at alongside many other factors may be treated differently from a score that automatically decides who gets an interview.
New York City Local Law 144
New York City was the first US jurisdiction to regulate AEDTs directly. The law applies to employers and employment agencies that use an AEDT to evaluate candidates or employees for jobs located in the city, with enforcement that began in July 2023.
- Bias audit: the tool must have an independent bias audit within one year before use, looking at selection or scoring rates by sex and race or ethnicity categories, including intersections.
- Public summary: a summary of the most recent audit results and the date the tool started being used must be published on the employer's website.
- Candidate notice: candidates and employees who live in the city must be told, in advance, that an AEDT will be used and which job qualifications and characteristics it will assess.
- Alternative process: the notice must explain how to request an alternative selection process or a reasonable accommodation, where available.
- Data information: information about the type of data collected, its source and the retention policy must be available.
- Penalties: violations can lead to civil penalties for each violation.
Other AI hiring laws to know
| Jurisdiction | What it covers in brief |
|---|---|
| Federal (US) | Title VII, the ADA and the ADEA apply to decisions made with AI tools. A tool that screens out a protected group without being job-related can create disparate impact liability, and employers remain responsible even if a vendor built it. |
| Illinois | The Artificial Intelligence Video Interview Act requires notice, an explanation of how AI evaluates video interviews and the candidate's consent, with limits on sharing and rules on deletion. |
| Colorado | A broader law on high-risk AI systems, including those used in employment decisions, with duties for developers and deployers being phased in. |
| European Union | The EU AI Act treats AI used in recruitment and candidate evaluation as high-risk, with obligations such as risk management, human oversight and transparency phasing in. |
What a bias audit looks at
A bias audit compares how a tool treats different groups. In its simplest form, it calculates the selection rate for each group (the share of candidates the tool advances) and divides it by the rate for the most selected group, producing an impact ratio. Large gaps prompt a closer look at whether the tool is measuring something genuinely related to the job.
- 1
Collect outcome data
Gather the tool's results for real candidates, with demographic data where it has been collected voluntarily.
- 2
Calculate rates by group
Work out selection or average scoring rates for each sex and race or ethnicity category, and their combinations.
- 3
Compare groups
Compute impact ratios against the most favored group.
- 4
Investigate gaps
Where a group is advanced much less often, check whether the criteria are job-related and whether the tool should change.
- 5
Repeat yearly
Audit again at least once a year, or after significant changes to the tool or how it is used.
AI hiring compliance checklist
- List every tool in your hiring process that scores, ranks or filters candidates.
- For each, record what it assesses, how its output is used and who makes the final decision.
- Keep a person accountable for every hiring decision; use AI to inform, not to decide alone.
- Make sure the criteria the tool scores are job-related and documented.
- Ask vendors for bias audit results, documentation and how their models are tested.
- Arrange independent bias audits where the law requires them, and publish summaries where required.
- Notify candidates in advance where required, explain what is assessed and how to request an alternative or accommodation.
- Review outcome data regularly for gaps between groups, even where no audit is required.
- Set data retention rules for candidate data and recordings.
How to tell candidates AI is used
Clear notice is required in some places and good practice everywhere. Candidates are more comfortable with AI in screening when they know what it does, what it does not do and who makes the decision.
- Say which step uses AI, for example scoring recorded answers or ranking applications.
- Say what it assesses, in terms of the job: the skills and qualifications, not vague traits.
- Explain that a person reviews the results and makes the hiring decision.
- Explain how to request an alternative process or an accommodation, and who to contact.
- Say how long candidate data and recordings are kept.
Best practices for using AI in screening
- Score against an explicit rubric you write, so you can explain why a candidate scored as they did.
- Use the same tool and settings for every candidate in a role.
- Let reviewers see the evidence behind a score, such as the answer or the test breakdown, not just the number.
- Treat low scores as a reason for a closer look, not an automatic rejection, especially near the cut-off.
- Tell candidates plainly that AI is part of the process.
springhire, for example, scores recorded interview answers with AI against a rubric the employer sets and shows the answer behind each score, while the hiring team makes the decision. Employers should still assess whether their own use of any such tool falls under laws like Local Law 144.
