In this article
AI in hiring is often described as a wild west. It is not. There are rules now, and they are more specific than most job seekers realise. Whether you are applying or hiring, it helps to know what the law actually says.
The short version
- New York City requires covered employers to audit automated hiring tools for bias, publish the results and notify candidates.
- The EEOC has said civil-rights law applies to algorithmic decisions the same way it applies to a human’s.
- “The algorithm did it” is not a legal defence.
New York City’s bias-audit law
Local Law 144, enforced since 5 July 2023, applies to employers and employment agencies in New York City that use an “automated employment decision tool” — software that substantially assists or replaces a human decision about who to hire or promote. If a tool is covered, the employer must:
- Have an annual independent bias audit of the tool.
- Publish a summary of the audit results.
- Notify candidates who live in New York City at least ten business days before the tool is used on them.
- Tell them what job qualifications and characteristics the tool assesses.
- Offer an alternative process or a reasonable accommodation on request.
Note what the law does and does not do. It does not ban AI in hiring. It requires disclosure and auditing. The theory is that a tool making decisions about people should be measured for unfair impact, just as a human decision would be.
What the EEOC says
In May 2023, the Equal Employment Opportunity Commission issued guidance making a simple point: existing civil-rights law does not stop applying because software is involved. If an automated tool causes a disparate impact based on race, colour, religion, sex or national origin, the employer can be liable under Title VII.
The guidance also says that using a tool does not transfer responsibility to the vendor. If the employer relies on the tool to make decisions, the employer owns the outcome. “The algorithm did it” is not a defence.
What this looks like in practice
Most large employers use an applicant tracking system. Jobscan detected one on 98.4% of Fortune 500 career sites in 2024. Many use keyword ranking and knockout questions, and a growing number add some form of AI-assisted screening. The level of automation varies enormously, from a simple keyword sort to models that rank candidates on predicted performance.
Even with all that machinery, the underlying problem is often not the model. Harvard Business School and Accenture found that 88% of employers admit exact-match criteria screen out qualified high-skill candidates. That is a design problem — a job description full of “required” skills and a filter built to enforce it — more than a machine-learning problem.
If you are applying
- In New York City, you can ask whether an automated tool is being used, and request an accommodation or an alternative process. Outside the city, you can still ask a recruiter; you may not get a detailed answer, but the question is legitimate.
- If you believe a knockout decision was wrong, you can ask for a human review. It does not always work, but it is a real route and sometimes it does.
- Keep your resume truthful and legible. The best defence against a bad filter is a record a human can verify in thirty seconds.
- Use human routes — a referral, a direct email, a short message to the hiring manager. They bypass more filters than any keyword trick, and they are the part of the process you actually control.
If you are hiring
- Audit the tool, and document what it does and what data it uses.
- Keep a human in the loop for any rejection that matters, especially ones that remove a candidate automatically.
- Separate “required” from “nice to have” in the job description. Most of the screening damage starts there.
- Track outcomes by group. If the tool is producing a skewed funnel, that is both a legal risk and a talent problem.
The rules are not perfect, and enforcement is uneven. But they establish a principle worth repeating: automation does not move the responsibility. A person still has to answer for the decision, and increasingly, that person has to be able to show their work.
Sources
Check your own resume against a real job ad.
Match score and two previews before you pay. $1 only if you like the tailored version.
Check my match score