AI Candidate Screening for Staffing and Recruiting Firms: Where a Human Still Has to Decide.

AI candidate screening for staffing agencies can parse, sort, and schedule. Here is where a recruiter still makes the call, and the hiring laws that shape it.

Where the line sits

The question for an agency owner is which screening steps you can hand to software without picking up a legal problem. Not all steps are equal. Sorting and scheduling? Clerical. Deciding who gets turned away is where the law gets involved.

Here is one way to divide the work:

  • The agent does it. Pull skills, experience and education from resumes into fields. Ask basic screening questions and log the answers. Offer interview times, confirm them, send reminders. Send "we received your application" messages.
  • The agent drafts, a recruiter confirms. A ranked shortlist, with the reasons shown. Rejection messages.
  • A recruiter only. Rejecting a candidate the agent screened out. Changing the criteria. Any candidate who has asked for an accommodation or raised a protected-class issue.

That split is a design choice, not a legal rule. It matches where the laws and lawsuits below keep ending up: the moment software, not a person, makes the call. Everything here is a business summary as of November 2026, not legal advice. Check the details with an employment lawyer in every state where you place people.

What an agent can take over

Most of what's published about these steps comes from recruiting-software vendors. Vendors such as Bullhorn describe the pattern: parse the resume, rank it against the job spec, generate screening questions.

Resume parsing. The agent turns a pile of resumes into fields and scores each one against what the job needs. What matters is how the score gets used. A score that helps a recruiter read faster is lower risk than one that cuts people. The Kistler case below shows the score itself can be challenged.

Screening questions. These are pass-or-fail basics: work authorization, a required license, shift availability. The agent asks them and records the answers. The decision about which questions are lawful and job-related, and what happens when a candidate fails one, stays with a person.

Scheduling and follow-up. The agent offers open slots from shared calendars, confirms them, sends reminders and reschedules. On its own, this step decides nothing about who gets an interview. Still, the wording and timing of messages to people who were screened out deserve a recruiter's eye.

Why staffing firms are a distinct case

An agency sits between the candidate, the software vendor and the client. The American Staffing Association said in September 2026 that a software vendor, the staffing firm and the client employer may all share exposure for a discriminatory algorithm. The same piece notes that a single platform can reach thousands of candidates a week.

The New York City rule below names employment agencies directly, so it applies to a recruiting firm itself, not just its clients, when the firm uses one of these tools for New York City hiring.

The rules, place by place

Each status below carries its date. The law is a patchwork, and it keeps shifting.

New York City. Local Law 144 lets an employer or employment agency use an automated employment decision tool only if the tool had a bias audit within the past year, the audit information is public and candidates received the required notices. The city's Department of Consumer and Worker Protection began enforcing it on July 5, 2023, and its guidance describes notice 10 business days before use. DLA Piper reports penalties of up to $1,500 per violation per day; the city's page does not state a penalty, so that figure rests on one law firm.

Little enforcement shows up in the record. A December 2025 audit by the New York State Comptroller found the city agency received two complaints about these tools in two years. Its own officials said they don't have the technical expertise to evaluate them. DLA Piper reads the audit as a sign of stronger enforcement to come. That is a forecast, not a fact.

Illinois. A law that took effect January 1, 2026 amends the state's Human Rights Act. According to DLA Piper and Thompson Hine, it bars AI use that discriminates against protected classes in hiring, bars ZIP codes as a stand-in for protected traits, requires notice to workers, and covers employers with even one Illinois employee. It doesn't require a bias audit. The state's Department of Human Rights is to write rules, so check whether they have been issued. A separate Illinois law covers AI analysis of recorded video interviews, with notice and consent requirements. A text-and-scheduling agent does not analyze video, so it does not trigger that one. A tool that scores video interviews would.

Colorado. The state's 2024 AI act never took effect. Lawmakers passed SB 26-189 in May 2026 to repeal it and put a new law in its place, and law firms such as Norton Rose Fulbright report it takes effect January 1, 2027. Employment is one of the areas it covers. Troutman Pepper Locke describes duties to give notice before use, explain an adverse outcome in plain language, and offer a chance for meaningful human review. Norton Rose says a trained person with authority to override the decision must be designated. That is the same basic shape as the split above: a named human who can overrule the software.

Federal. The EEOC removed its AI guidance on adverse impact from its site in early 2025, and an April 2025 executive order told agencies to deprioritize disparate-impact enforcement. Removing guidance does not change Title VII, the age discrimination law or the disability law. Private plaintiffs can still bring disparate-impact claims, as Maynard Nexsen summarizes. A later order set up a Justice Department task force to challenge state AI laws. As of an April 2026 analysis, it had filed no suit, and firms were advising employers to keep following state laws.

Cases worth watching

A plain filter in what has been called the EEOC's first AI-related settlement. In 2023, the EEOC settled with iTutorGroup for $365,000 over screening software that allegedly rejected women 55 and older and men 60 and older automatically. iTutorGroup denied the allegations. The mechanism was a cutoff based on birth date. An ordinary rule, nothing exotic. A screening question that works the same way could carry the same risk.

Mobley v. Workday. A federal court in California allowed claims to proceed on the theory that a software vendor can be an "agent" of the employers it serves, and that can bring it within the legal definition of an employer. In May 2025 the court gave preliminary certification to an age discrimination collective, which is conditional, not a final class. The case is pending. There's no ruling on liability. A staffing firm acting on a client's behalf fits the same shape.

Kistler v. Eightfold AI. Filed in January 2026, this suit alleges that an AI vendor's candidate scores are consumer reports under federal credit-reporting law, and that lower-ranked candidates are often discarded before a person sees them, per Epstein Becker Green. These are allegations only, and no class has been certified. The risk it describes: a scoring layer nobody reviews.

Guardrails that follow

These come from the rules and cases above. They're a design approach, not a compliance checklist.

  • Put a named recruiter on every rejection. The software sorts, and a person decides.
  • Keep fewer fields. No date of birth for screening. No ZIP code as a filter.
  • Skip video and voice analysis unless you have worked through the Illinois video law with counsel.
  • Keep a log of what the agent saw, scored and sent, held for the longest period required in any state where you place candidates. The American Staffing Association recommends this.
  • Compare outcomes. Check selection rates by group on a sample. One compliance blog points to the four-fifths rule for that check.
  • Tell candidates when a law requires it, as in New York City and Illinois.

What a build looks like for a recruiting firm

Tarsus builds working systems inside a client's tools, and the design rule for recruiting firms is the one above: the agent ranks applicants and schedules interviews, and a person makes every decision. The shortlist goes to a recruiter to review. You can see how that looks on the HR and recruiting firms page and in the hiring solution.

The same handoff thinking applies in other professions. See how it plays out for accounting firms, where a CPA keeps the judgment calls, and for dental practices, where emergencies and clinical questions stay with staff.

Keep reading.

All posts