Video Interview AI and the New York Bias Audit
If your video-interview tool scores candidates, it likely needs an annual bias audit and candidate notices. Here's the operational path.
By Rovaryn Digital · · 8 min read

When a Video Interview Tool Becomes an AEDT
Your recruiting team just rolled out a video-interview platform that ranks candidates by "communication score" before a human ever watches the clip. Nobody in the room called it an AI system. Nobody flagged it for legal review. It was procured as an interview scheduler with a nice-to-have scoring feature, and it went live on a Tuesday.
Three weeks later, someone in HR asks the question that should have come first: does this tool decide who moves forward, or does it just record the interview? If the answer is anywhere close to "it ranks them," you may already be out of compliance with a New York City law that carries daily penalties and a hard notice deadline you can't retroactively satisfy.
This is the situation a lot of HR directors find themselves in with video-interview AI specifically, because the tool doesn't look like the "resume screener" everyone pictures when they think of AEDTs. By the end of this article, you'll be able to determine whether your video-interview tool triggers Local Law 144, and what the candidate notice actually has to contain before your next interview slot goes out.
Why Video Interview AI Triggers a New York Bias Audit
Local Law 144 doesn't regulate video interviews. It regulates a specific function: a computational process that issues a simplified output — a score, classification, or recommendation — used to substantially assist or replace a discretionary employment decision. That definition, drawn from machine learning, statistical modeling, data analytics, or AI, is deliberately tool-agnostic (Perkins Coie, 2023). A resume parser fits it. A ranking algorithm fits it. And a video-interview platform that scores tone, word choice, facial movement, or "engagement" against a model and hands your recruiter a number fits it too.
The mechanism that matters for a video interview ai bias audit new york determination is simple: does the platform reduce the candidate to a score or rank before a person exercises independent judgment? If a recruiter can look at every transcript with no score attached and reach their own conclusion, you're closer to "just a recording tool." If the platform sorts candidates into tiers, flags top performers, or auto-advances anyone above a threshold, you're almost certainly running an AEDT — and Local Law 144 has applied to AEDTs used on NYC candidates and employees since it took effect January 1, 2023, with enforcement beginning July 5, 2023 (Epstein Becker Green / Workforce Bulletin, 2023).
Worth saying plainly here: this isn't a call WorkforceNewYork makes for you, and it isn't legal advice. It's an operational read of how the statute's language maps onto a scoring feature. If the classification is close, confirm it with DCWP guidance or outside counsel before you build your compliance calendar around it. For the broader classification test — including tools that don't obviously look like "AI" — see what is an automated employment decision tool and is my hiring tool an AEDT.
The Three Obligations Once Your Tool Qualifies
Once a video interview scoring AEDT is confirmed in scope, Local Law 144 attaches three separate obligations, and video-interview tools tend to trip on all three because the workflow moves fast (Crowell & Moring LLP, 2023; Epstein Becker Green, 2023):
- An annual independent bias audit, performed by an auditor with no financial or employment relationship to you or the tool's vendor. Your team doesn't run this audit and neither does WorkforceNewYork — you engage an independent auditor, and the audit reviews scoring outcomes across protected categories.
- A public summary of the most recent audit results, posted on your website along with the date the AEDT was first used and the date of the summary.
- Candidate notice at least 10 business days before the tool is used, with a path to request an alternative process or accommodation.
The four-fifths rule is the analytical backbone auditors apply inside that bias audit: a selection rate for any group below 80% of the rate for the highest-selected group can indicate adverse impact (EEOC Uniform Guidelines, via Assessment Systems, 2024). As a worked example — not a claim about your data — if your highest-scoring group advances at a 50% rate, any group advancing at less than 40% (80% of 50%) would flag for closer review under that rule. That's the kind of pattern an independent auditor is checking for in a video-scoring context, where subtle correlations between speech patterns, accent, or even camera lighting and downstream scores can hide inside an "engagement" metric nobody defined precisely.
The penalties for skipping any of the three obligations run up to $500 for a first violation, escalating to $500–$1,500 for each subsequent violation, and they accrue per violation per day (Office of the NY State Comptroller, 2025). A video-interview rollout that runs for two months without a posted summary or a compliant notice isn't a single infraction — it's a running clock.
What Must an AEDT Candidate Notice Include
This is the piece video-interview deployments get wrong most often, because the notice has to go out before the tool touches a real candidate, not after the interview is scheduled and definitely not after the score is generated.
At minimum, a compliant notice needs to tell the candidate:
- That an AEDT will be used to assess them in the process
- What the tool is assessing (e.g., "your recorded video responses will be scored for communication and job-fit indicators")
- The candidate's right to request an alternative selection process or a reasonable accommodation
- Where to find the AEDT's job qualifications and characteristics if the employer discloses them
That notice has to land at least 10 business days before the video-interview tool is used on the candidate — not 10 business days before the interview is scheduled, and not 10 business days before the offer decision. If your ATS auto-sends the interview link the same day a candidate applies, and the scoring engine runs the moment the recording uploads, there may be no 10-business-day window left to satisfy. That's a scheduling and system-design problem as much as a compliance one. The full mechanics of counting the window, handling weekends and holidays, and documenting delivery live at the AEDT 10-business-day notice guide.
Candidate Consent vs. Notice: Why They're Not the Same Thing
A recurring mix-up: teams treat a broad "we may use AI tools in our hiring process" consent checkbox, buried in an application form, as satisfying the notice requirement. It doesn't. Consent to be recorded or evaluated is a different legal concept from the specific, timed, AEDT-focused notice Local Law 144 requires. A consent checkbox collected at application time, with no reference to the specific tool, no accommodation pathway, and no 10-business-day runway, does not stand in for a compliant notice — even if your legal team is comfortable with it for other purposes.
This distinction matters most for video interviews specifically because recording consent is already a familiar, separate compliance touchpoint (often tied to state recording laws), and it's easy to assume one form covers both purposes. It doesn't. Track them as two different documents with two different triggers. The detailed breakdown of where consent language ends and AEDT notice language begins is at candidate consent and AI resume screening.
Where Video Scoring Tools Get This Wrong
Independent research on published Local Law 144 disclosures found that among 391 employers reviewed, only 18 had posted an audit report and only 13 had posted a transparency notice (ACM FAccT, Wright & Muenster et al., 2024).
That gap isn't unique to video-interview tools, but video scoring sits inside it disproportionately, because the tool often arrives through a vendor relationship owned by talent acquisition or IT rather than HR compliance — nobody downstream realizes the scoring feature exists until a candidate asks a question the recruiter can't answer. The operational fix isn't a bigger legal review cycle. It's an inventory habit: every time a new interview, screening, or ranking feature ships — even a "pilot" — someone logs whether it scores candidates, and if it does, the notice and audit clocks start immediately, not after the pilot proves out.
For the full statutory walkthrough of how Local Law 144 obligations interact — audit cadence, public posting, and notice together — see the NYC Local Law 144 compliance guide.
Your First Action Item
Before your next video-interview cycle goes live, confirm two things: whether the scoring feature makes your tool an AEDT, and whether your current notice language actually names the tool, the assessment, the accommodation path, and the 10-business-day window. If any of those pieces are missing, the Candidate Notice Operations Workbook — Screener / Video / Ranking Notice Sets gives you pre-built notice templates and a tracking log split by tool type, so a video-scoring rollout doesn't get the same generic notice as a resume screener when the timing and disclosure content actually differ. Download it, adapt the language to your specific tool, and route it to counsel for a final look before it goes to candidates — the workbook runs the documentation operation; it doesn't replace that review.
Related guides
- AEDT Inventory & Scope
The NYC Residency AEDT Notice Trigger, Requisition by Requisition
The notice trigger is a per-requisition decision. Here's a repeatable way to make it and document it.
Rovaryn Digital · · 6 min read
- AEDT Inventory & Scope
Does Local Law 144 Apply to Candidates Outside NYC?
Not every candidate triggers a notice. The residency and office-association tests that decide who's in and who's out.
Rovaryn Digital · · 7 min read
- AEDT Inventory & Scope
Local Law 144, Remote Workers, and NYC Candidates
Remote roles muddy the LL144 trigger. Here's how residency and NYC office association decide whether a notice is owed.
Rovaryn Digital · · 8 min read


