The line the law draws through screening
Why does the brief carry no score, no rank, no reject flag? Because the moment a tool decides — scores, ranks, or rejects a candidate — you've crossed from drafting into the single most regulated act in AI-era HR. The map, as of this writing:
- NYC Local Law 144 — enforced since July 5, 2023. Employers using automated employment decision tools for NYC candidates must commission an independent annual bias audit, publicly post a summary of the results, and give candidates at least 10 business days' notice before the tool is used on them. Penalties run $500 to $1,500 per day.
- Illinois — the AI Video Interview Act requires that applicants be notified that AI will analyze their video interview, given an explanation of how it works, and asked for consent before it runs. A broader Illinois law effective January 2026 prohibits AI use with discriminatory effect across employment decisions generally.
- Colorado — the state's AI law (as amended, effective January 1, 2027) targets automated decision-making that materially influences major employment decisions: notice to the person, a structured adverse-action process with human review, and three-year record retention. It has already been amended once; anyone relying on it should re-verify at decision time.
- Mobley v. Workday — the case that makes this everyone's problem. Filed 2023 in the Northern District of California, claiming an AI screening system discriminated by race, disability, and age. In May 2025 the court granted preliminary certification of an age-discrimination collective; opt-in notices went out starting early 2026; Workday told the court it had processed 1.1 billion rejected applications. The EEOC argued in an amicus brief that the vendor can be covered under Title VII, the ADEA, and the ADA. Translation: "the tool did it" identifies a co-defendant, not a defense — and the liability theory doesn't wait for your city to pass a law.
One more thing a compliance-minded specialist should clock: the federal posture went the other direction. The EEOC's 2023 AI guidance came off its website in early 2025, and an April 2025 executive order steered agencies away from disparate-impact enforcement. If that reads like relief, read the docket again — enforcement didn't shrink, it moved to state statutes and private collective actions, which are harder to predict and don't answer to elections the same way.
So the studio rule generalizes the strictest regime, everywhere, regardless of where you hire:
No tool scores, ranks, or rejects a candidate. If your ATS has those features, they stay off until candidates get notice, the tool has an audit posture, adverse outcomes get human review, and the decision records are retained.
That last clause — decision records retained — is where lesson three picks up. The brief you can audit is half the defense; the decision trail is the other half.