AI Is Screening Your Candidates. Can You Prove It's Fair?

Written By Ivan Perry

Most companies using AI to screen candidates aren't asking whether it's legal. They're assuming it is, because everyone else is doing it too. That assumption is getting more expensive by the month.

The Regulatory Floor Is Already Set

NYC Local Law 144 requires an annual bias audit from an independent auditor, a public summary of the results, and at least 10 business days notice to candidates before an automated employment decision tool, or AEDT, factors into their evaluation. Illinois HB 3773, effective January 2026, makes it unlawful to use AI that produces a discriminatory effect in hiring, and it specifically bars using ZIP code as a stand in for protected class. Colorado's original AI law was repealed and reenacted in May 2026 as the Automated Decision-Making Technology Act. It covers any tool used to make, guide, or assist a consequential employment decision, and requires employers to notify candidates before the tool is used and explain its role within 30 days of an adverse decision. Those obligations take effect January 1, 2027. It is not a far reach to claim that other states will soon follow suit.

The EU AI Act, formally Regulation 2024/1689, classifies AI used in recruitment and candidate screening as “high risk” per their classification system. That covers tools that screen resumes, rank applicants, or run targeted job ads, and it brings mandatory risk assessments, bias testing, human oversight, and transparency disclosures with it.

None of these laws ban AI screening. They regulate the proof burden that comes with it. The real question for most companies is not whether they can use AI. It's whether they can produce the audit trail if someone challenges the outcome.

Liability Runs Wider Than the State List

AI used in hiring is still a selection procedure under Title VII. That rests on the Uniform Guidelines on Employee Selection Procedures, a federal regulation in place since 1978, not on the technical guidance the EEOC published in 2023 explaining how it applies to AI specifically. That guidance was pulled from the EEOC's website in January 2025. The underlying regulation was not repealed, and employers stay liable for discriminatory outcomes whether the tool was built internally or bought from a vendor.

This isn't theoretical. The EEOC has already settled AI hiring discrimination claims, and vendor liability is being tested further right now in Mobley v. Workday, where the claim is that an AI vendor can be held liable directly, not just the employer using the tool, under a theory that the vendor acted as the employer's agent in making the hiring recommendation. It's the same exposure a retained search firm or staffing firm takes on the moment it uses AI to source, screen, or shortlist on a client's behalf.

What Proving a Fair Process Actually Requires

The standard here is specific, not conceptual. The EEOC's four fifths rule is the trigger. An impact ratio below 0.80 between any demographic group and the most selected group generally signals potential adverse impact. That ratio gets calculated by sex, race, ethnicity, and the intersections where the data allows it. A published audit showing a ratio below that line, with no documented remediation, is not a compliance footnote. It's evidence a plaintiff's attorney can use.

This isn't new territory for me. In every TA function I've led, building a screening process that could stand up to scrutiny wasn't optional. It mattered for the audit, and it mattered for the candidate on the other side of it, whether they moved forward or not. Internal or external, a fair process is part of the experience a candidate remembers.

Five things to put in place now

None of these require waiting on a specific state's effective date.

  • Inventory every tool that screens, scores, or ranks candidates, including the ones buried inside a larger HR platform nobody thinks of as AI.

  • Get documentation from vendors on training data, bias testing methodology, and who owns audit cost and liability in the contract.

  • Build a real human review step before a rejection becomes final, not a rubber stamp on the algorithm's output.

  • Keep the audit trail for individual automated decisions, not just the aggregate bias numbers.

  • Update candidate facing notices to disclose AI use before it happens, not after someone files a complaint.

Where This Framing Falls Short

These laws don't ask companies to explain individual rejections to individual candidates. They ask for aggregate statistical proof that the tool doesn't produce disparate outcomes across protected groups, audited independently and posted publicly on a repeating schedule. That's a heavier lift than a well written rejection email. Treating the two as the same thing is how companies end up unprepared.

This Isn't Just a Legal Problem or Just an HR Problem

General Counsel should not run these audits alone, and HR should not run them alone either. The audit sits at the intersection of legal exposure, HR process design, and data governance. When one function owns it in isolation, something gets missed. GC without HR misses how the tool actually gets used day to day, where managers create workarounds, or how scoring gets weighted in practice. HR without GC misses how the audit trail holds up under legal challenge or regulatory scrutiny.

The audit itself is only the first step. Once findings exist, someone has to communicate the plan back to the business, not just file the report. Business leaders and hiring managers need to know the policy exists, why it exists, and what changes for them day to day. Skip that step, and a compliant policy on paper does nothing to stop a hiring manager from working around the tool or a business leader from ignoring the process to fill a role faster. That gap between policy and practice is exactly what a bias audit or a discrimination claim will expose.

Process and policy is not overhead. It's what keeps hiring managers and business leaders inside the guardrails instead of making ad hoc calls that create exposure. The upfront cost of building the right foundation, cross functional buy-in, clear communication, documented process, is real. It pays back the first time a candidate challenges a decision, the first time an auditor asks for records, or the first time a manager tries to skip a step. Companies that treat this as a governance investment rather than a compliance checkbox end up with a better candidate experience on both sides, external hires and internal moves and promotions, because the criteria stay consistent and defensible either way.

Retained Search Firms and Staffing Firms Carry the Same Exposure

None of this stops at the employer's front door. Retained search firms and staffing firms that use AI to source, screen, or shortlist candidates on a client's behalf carry the same exposure the client does, and in some cases more, since the firm is acting as an agent making hiring recommendations for the client. That is the same agent theory now being litigated in Mobley v. Workday. A firm that can't show its own screening process is fair and documented puts every client relationship at risk, not just the search in question.

I've sat on both sides of this, in-house building and running TA functions, and inside retained search delivering searches for clients. That combination isn't common, and it's why I look at AI-driven screening from both directions, the exposure a company takes on by trusting a vendor's tool, and the exposure a search firm takes on by using one without vetting it first. Retained search firms aren't exempt from what NYC, Illinois, and Colorado require, and clients should be asking their search partners the same audit and documentation questions they'd ask any automated employment decision tool (AEDT) vendor.

Enforcement Is Tightening, Not Loosening

A December 2025 audit of NYC's own enforcement agency found major gaps, including most test complaints never reaching the department responsible for reviewing them. That's not a reason to relax. It's the reason enforcement gets tighter through the rest of 2026, not looser. AI screening isn't the problem. An AI screening process nobody can defend is.

How I Help With This

This is the work I do at ADVON Leadership Advisors. Talent acquisition advisory built on both sides of the table, in-house TA leadership and retained executive search, grounded in the research-intensive discipline I learned as an intelligence officer in the US Air Force. I help organizations inventory the AI already sitting inside their hiring stack, stand up the audit and documentation process before a regulator or a plaintiff's attorney asks for it, and build the internal communication plan that keeps hiring managers and business leaders following the policy instead of working around it. For organizations that use retained search, I also help vet whether search partners' own AI practices create exposure the client hasn't accounted for.

If your hiring process leans on AI and you haven't stress tested it, or if you want a second set of eyes on how your search partners screen on your behalf, reach out through www.advonleadershipadvisors.com. Outright avoiding AI or assuming employees aren’t using it, is not a strategy.

Sources

Primary sources for the regulations and cases referenced above, all government or official EU sources.

Ivan Perry

Ivan has held HR leadership roles in global tech organizations, advising executives, CHROs, and Talent Acquisition leaders on talent strategy, performance, and leadership consulting. He brings an evidence-based and practical approach that supports clear decisions and measurable outcomes.

He focuses on executive assessments, leadership consulting, talent strategy, coaching, and inclusive workforce solutions for mission-driven organizations. He supports executive effectiveness through leadership development, executive selection, onboarding, and career development.

He incorporates Hogan Assessments, informed by his experience leading global talent functions, to strengthen client decision-making through structured insight, clear judgment, and accountable follow through.

His work is informed by earlier experience as a Partner in retained executive search and as an Air Force Intelligence Officer for the 18th Wing and 33d Rescue Squadron, bringing structured assessment and pragmatic execution to senior level talent decisions.

Previous

Hypergrowth Doesn't Build Leaders. It Reveals Them.
Next

The Build or Buy Question Most Mid-Sized Organizations Answer Too Late