OVI is the AI-native ATS where you chat to hire. Two AI agents — Sora for sourcing and Milo for screening — find candidates, score every CV, and run AI interviews, all controlled from a single chat.

What does "chat to hire" mean?

Chat to hire is OVI's AI-native hiring workflow: instead of clicking through a traditional ATS, you describe what you need in chat and OVI's AI agents source candidates, screen applicants, run AI interviews, and send outreach for you.

What is an AI-native ATS?

An AI-native ATS is an applicant tracking system built around AI agents rather than forms and pipelines. OVI's AI chat ATS lets recruiters source, screen, interview, and follow up through conversation — the agents do the work and surface only the candidates worth meeting.

Who are Sora and Milo?

Sora is OVI's AI sourcing agent — she scans LinkedIn, the open web, and your ATS history to find best-fit candidates and reach out in your name. Milo is OVI's AI screening agent — he scores every applicant against your rubric and surfaces a ranked shortlist.

How do OVI's AI interviews work?

OVI runs async AI interviews for every shortlisted candidate and returns scores, transcripts, and recordings — so your team only meets the people worth meeting.

Your AI Hiring Tool Has a Bias Audit. Regulators Just Proved It Does Not Matter.

2026-06-19

Your AI Hiring Tool Has a Bias Audit. Regulators Just Proved It Does Not Matter.

Current date (UTC): 2026-05-01
Current time (UTC): 09:33

Of the hundreds of complaints New Yorkers filed about automated hiring tools, three out of four never reached the agency responsible for enforcing the law. The complaints were routed to 311, logged, and lost in a bureaucratic gap between city departments. When state auditors finally examined the same set of employers the city had cleared, they found seventeen potential violations the city's own investigators had missed.

That is not a rounding error. It is a systemic failure — and it means the bias audit sitting in your vendor's compliance folder may be worth less than the PDF it is printed on.

Part 1: The Enforcement Collapse

New York City's Local Law 144, which took effect in July 2023, was the first US regulation requiring employers to conduct independent bias audits before using AI in hiring (NY State Comptroller, Dec 2025). The law was supposed to be a model. A December 2025 audit by the New York State Comptroller told a different story (NY State Comptroller, Dec 2025).

The Comptroller's office found that 75 percent of complaints submitted via New York's 311 system were misrouted and never forwarded to the Department of Consumer and Worker Protection (DCWP), the agency tasked with enforcement. Of the complaints that did reach DCWP, investigators examined 32 firms — and found just one violation. State auditors reviewed the same sample and identified 17 potential violations the city had overlooked.

The gap was not subtle. DCWP lacked the technical expertise and investigative processes to evaluate whether an employer's bias audit was methodologically sound. In practice, the agency accepted vendor-produced audit reports at face value without scrutinizing the underlying data or methodology.

For employers, the takeaway is uncomfortable: passing a LL 144 audit does not mean your hiring process is compliant. It may only mean no one looked closely enough.

Part 2: Auditing the Audits

If the enforcement side is broken, the audits themselves are not faring much better.

A landmark study presented at ACM FAccT 2025 analyzed 116 publicly available LL 144 bias audits — the largest systematic review of its kind. The findings were damning (ACM FAccT 2025).

Researchers found widespread methodological problems across the audit vendor market. Many audits relied on incomplete demographic data, used opaque aggregation methods that masked disparities, and in some cases substituted synthetic or proxy test data when actual applicant data was unavailable. When the researchers accounted for the missing data, the four-fifths rule — the standard EEOC threshold for identifying adverse impact — was likely violated in cases that had otherwise received clean reports.

The audit vendor market itself is highly concentrated. Just three firms — Holistic AI (20%), DCI Consulting (20%), and BABL AI (16%) — conducted more than half of all published audits (ACM FAccT 2025). While BABL AI has developed a Criterion Audit Framework modeled on financial auditing standards to bring greater rigor to the process (BABL AI), the FAccT study suggests such rigor is the exception rather than the norm across the industry.

As DLA Piper warned in January 2026, the Comptroller's findings "signal increased risk for employers" who rely on vendor audits as a compliance safe harbor (DLA Piper, Jan 2026). An audit that does not withstand regulatory scrutiny is not a shield — it is a liability.

Part 3: The Regulatory Patchwork Is Getting Bigger

New York's enforcement problems would matter less if LL 144 were an isolated experiment. It is not.

The US multi-state patchwork is expanding. Illinois enacted its AI in Employment law effective January 1, 2026. Colorado's SB 24-205 follows on June 30, 2026 (Holland & Knight, Mar 2025). At the federal level, no unified AI hiring law has emerged — but Title VII and ADA protections still apply to AI-mediated employment decisions (Holland & Knight, Mar 2025). Employers operating across state lines now face overlapping, sometimes contradictory obligations with no unified federal framework.

The EU AI Act raises the bar further. Under Annex III, AI systems used for recruitment and selection are classified as high-risk (EU AI Act Annex III). Core obligations take effect on August 2, 2026 (though a deferral to December 2027 was proposed in November 2025). Since February 2025, one provision is already active: emotion recognition in job interviews is banned in the EU (Octagon People, Apr 2026).

For multinational employers, this means AI hiring tools must satisfy both US state-level audit requirements and EU high-risk system obligations — transparency, human oversight, data governance, and conformity assessments — simultaneously. The cost of non-compliance is not just fines. It is litigation risk, reputational damage, and the operational disruption of having to pull tools mid-cycle.

Research from Warden AI underscores the scale of the problem: bias in AI-powered talent acquisition remains prevalent, and most organizations lack the internal capabilities to detect it independently (Warden AI).

What HR Leaders Must Do Now

The audit-and-forget era is over. Here is what needs to change:

1. Stop treating vendor audits as compliance endpoints. A bias audit is a starting point, not a certificate of compliance. The Comptroller's findings prove that even the regulator charged with enforcement cannot reliably evaluate these audits. Your internal team needs to understand what your audit actually tested, what data it used, and what it excluded.

2. Build an internal review layer. Do not outsource your entire compliance posture to a single vendor report. Establish an internal process — or engage independent counsel — to stress-test your audit's methodology against the FAccT study's findings. Ask: does your audit rely on complete demographic data? Does it use the four-fifths rule correctly? Are results aggregated in ways that could mask disparities?

3. Track the regulatory calendar. Key dates:

Illinois AI Employment Law: January 1, 2026 (in effect)
Colorado SB 24-205: June 30, 2026
EU AI Act core obligations for high-risk AI: August 2, 2026 (possible deferral to December 2027)
EU emotion recognition ban in interviews: February 2025 (in effect)

4. Prepare for EU AI Act obligations now. If your organization uses AI hiring tools with EU-based candidates, start mapping your systems against Annex III requirements. The August 2026 deadline leaves limited runway, and conformity assessments require documentation that takes months to prepare.

5. Document everything. In a fragmented enforcement landscape, the best defense is a clear paper trail showing you evaluated risks, questioned your vendors, and made informed decisions. The absence of federal enforcement does not mean the absence of legal exposure — state attorneys general, private plaintiffs, and the EEOC's existing Title VII authority remain active vectors.

The regulatory framework for AI in hiring is fracturing faster than most compliance teams can track. The organizations that treat this as a governance challenge — not just a vendor checkbox — will be the ones that avoid the next enforcement action.

Sources:

What did the NY State Comptroller audit find about NYC LL 144 enforcement?

75% of complaints submitted via 311 were misrouted and never reached the DCWP. State auditors reviewing the same employer sample found 17 potential violations that city investigators had overlooked.

Are AI hiring bias audits legally sufficient for compliance?

Not necessarily. An ACM FAccT 2025 study of 116 LL 144 audits found widespread methodological problems — incomplete demographic data, opaque aggregation, and synthetic proxy data — meaning a clean audit report may still mask adverse impact violations.

What key AI hiring compliance deadlines should HR leaders track in 2026?

Illinois AI Employment Law (Jan 1, 2026 — in effect), Colorado SB 24-205 (June 30, 2026), EU AI Act high-risk AI obligations (August 2, 2026, possible deferral to Dec 2027), and the EU emotion recognition ban in interviews (Feb 2025 — already in effect).