Did the Eightfold Matching Model pass its 2026 bias audit?

Yes. The Eightfold Matching Model passed all three sections of the 2026 independent bias audit conducted by BABL AI Inc.: disparate impact, governance, and risk assessment. The overall result is PASS.

What is NYC Local Law 144 and does Eightfold comply?

New York City Local Law No. 144 of 2021 requires yearly bias audits for automated employment decision tools used in hiring or promotion. The audit must assess disparate impact across race/ethnicity and gender, be conducted by an independent auditor, and the summary of results must be made publicly available. Eightfold AI conducts this audit annually and publishes full results publicly.

What were the bias audit results for gender?

Female candidates showed a scoring rate of 62.8% (impact ratio 1.000, reference group). Male candidates showed a scoring rate of 60.5% (impact ratio 0.962). Both groups are well above the Four-Fifths Rule threshold of 0.80. No gender group showed adverse impact across 23.8 million assessed candidates.

What were the bias audit results by race and ethnicity?

All seven race/ethnicity groups exceeded the Four-Fifths Rule threshold. Hispanic or Latino candidates had the highest scoring rate at 67.2% (reference group, impact ratio 1.000). Asian candidates had the lowest scoring rate at 63.0% (impact ratio 0.938), still well above the 0.80 threshold. All groups were assessed across 15.4 million candidates with known race/ethnicity.

Does this audit certify that the Eightfold Matching Model is bias-free?

No. The audit does not certify that the Eightfold Matching Model is bias-free — no audit can make that claim. This audit was designed to satisfy the requirements of NYC Local Law 144, assessing disparate impact across gender and race/ethnicity. It is not intended to demonstrate compliance with any legislation other than the NYC AEDT law.

How does Eightfold's AI support hiring decisions?

The Eightfold Matching Model evaluates a candidate's skills and experience relative to a specific role, producing a match score that helps organizations make more informed talent decisions. The model supports human judgment — it does not replace it. Final decisions in any hiring or promotion process remain with the people using the platform.

How often does Eightfold AI conduct bias audits?

Eightfold AI conducts independent bias audits annually, in compliance with NYC Local Law 144. Full results are published publicly each year at the same level of detail.

Responsible AI | Transparency

AI Interviewer, audited for bias. Published in full.

Q: Who conducted Eightfold's 2026 bias audit?

The audit was conducted by BABL AI Inc., an independent AI auditing firm based in Iowa City, Iowa. Lead auditors are ForHumanity Certified Auditors under the NYC AEDT Bias Audit standard. The audit was signed by Dinah Rabe, Lead Auditor, on March 26, 2026.

AI Interviewer was independently audited for disparate impact under New York City Local Law 144. It passed. Here is every group, every score, and every limitation, including where the data is synthetic.

This page presents the results of the 2026 bias audit of AI Interviewer, conducted by BABL AI Inc. The audit assessed disparate impact across gender and race/ethnicity, as required by New York City Local Law No. 144 of 2021, along with internal governance and risk assessment. We even went beyond the requirements of the law and also assessed disparate impact across age. AI Interviewer passed all three audited sections.

Prepared by BABL AI Inc. | Audited June 29, 2026 | Signed: July 28, 2026

Audit Results · NYC Local Law 144

Audit conclusions

Three sections audited. Three passing opinions.

Disparate Impact

PASS

Governance

PASS

Risk Assessment

PASS

Overall

PASS

The audit was conducted by BABL AI Inc., an independent auditing firm whose lead auditors are ForHumanity Certified under the NYC AEDT Bias Audit standard. BABL AI independence conforms to the ForHumanity and Sarbanes-Oxley definitions. Fees are unrelated to the opinion rendered.

The system

What AI Interviewer does

AI Interviewer conducts AI-powered interviews. Given a job description, the AI Interviewer model generates a list of interview questions and evaluation criteria for the role for a customer’s review, revision, and/or approval. If the model determines that a candidate did not fully answer a question, it asks a follow-up so the candidate can give a complete answer.

After the interview completes, the model helps assess the candidate against the customer’s defined evaluation criteria, to help inform recruiters’ and hiring managers’ hiring decisions.

Auditor

BABL AI Inc.

Audit date

June 29, 2026

Signed

July 8, 2026

Testing conducted by

Eightfold

Date of most recent testing

May 2026

Data timeframe (gender and age)

January 2025 to December 2025

User-configurable settings that affect output

Job descriptions, interview questionnaires, scoring criteria

Applicable law

NYC Local Law 144

Methodology

How scoring rates and impact ratios work.

The audit used the scoring rate method – the proportion of candidates within a demographic group who scored at or above the overall median score of the full population. Impact ratios are calculated by dividing each group’s scoring rate by the scoring rate of the highest-scoring group. Under the federal Four-Fifths Rule (UGESP, 1978), an impact ratio below 0.80 is generally regarded as evidence of adverse impact.

In plain terms: male candidates scored at or above the median 0.754 of the time and female candidates 0.747 of the time, giving female candidates an impact ratio of 0.990, well above the 0.80 threshold. All groups in this audit remained above 0.80.

Data Transparency

Where these numbers come from.

Gender

Data from real historical use, with self-reported gender labels.

Age

Data from real historical use, with inferred age labels.

Race/ethnicity and intersectional

Synthetic data. Because insufficient real-world data was available for meaningful race/ethnicity and intersectional testing, a large language model was instructed to act as a candidate and conduct synthetic interviews across 12 candidate personas. Demographic data was incorporated through candidate names generated for each demographic group.

DISPARATE IMPACT · GENDER

Gender scoring rates, from real interviews.

Real historical usage data. Gender labels are self-reported.

Group	N applicants	Scoring rate	Impact ratio
Male	1,250	0.754	1.000 — reference
Female	395	0.747	0.990

An additional 1,235 applicants with an unknown gender category were not included in this calculation, a reflection of real-world self-declaration gaps.

DISPARATE IMPACT · RACE / ETHNICITY

Race and ethnicity scoring rates across seven groups.

Synthetic test data.

Group	N applicants	Scoring rate	Impact ratio
Black or African American	238	0.676	1.000 — reference
Two or more races	238	0.672	0.994
Native Hawaiian or Pacific Islander	238	0.664	0.981
Asian	238	0.660	0.975
Hispanic or Latino	238	0.660	0.975
White	238	0.655	0.969
Native American or Alaskan Native	238	0.651	0.963

All seven race/ethnicity groups showed impact ratios above 0.80.

Disparate Impact · Age

Age scoring rates, from real interviews.

Real historical usage data. Age labels are inferred.

Group	N applicants	Scoring rate	Impact ratio
Above 40	121	0.802	1.000 — reference
Below 40	2,759	0.723	0.902

Both age groups showed impact ratios above 0.80.

Disparate Impact · Intersectional

Gender and race/ethnicity combined.

Synthetic test data.

New York City Local Law 144 requires intersectional analysis of every combination of gender and race/ethnicity. The gender data here does not refer to the same candidates as the gender table above, and is synthetically constructed. The reference group is Black or African American Male.

Male candidates

Group	N applicants	Scoring rate	Impact ratio
Black or African American	119	0.689	1.000 — reference
Native Hawaiian or Pacific Islander	119	0.681	0.988
Two or more races	119	0.672	0.976
Asian	119	0.664	0.963
Hispanic or Latino	119	0.664	0.963
White	119	0.655	0.951
Native American or Alaskan Native	119	0.647	0.939

Female candidates

Group	N applicants	Scoring rate	Impact ratio
Black or African American	119	0.664	0.963
Native Hawaiian or Pacific Islander	119	0.647	0.939
Two or more races	119	0.672	0.976
Asian	119	0.655	0.951
Hispanic or Latino	119	0.655	0.951
White	119	0.655	0.951
Native American or Alaskan Native	119	0.655	0.951

All intersectional groups showed impact ratios above 0.80. The lowest observed ratio was 0.939.

Governance · Audit finding: Pass

Who owns fairness at Eightfold AI.

Governance of bias and fairness risk is managed by a cross-functional Responsible AI working group that includes the Chief AI Compliance Officer along with product, engineering, legal, and security representatives. The audit confirmed that an accountable party is identified, that its duties are clearly defined, and that those duties were carried out prior to the audit.

Accountable party: Legal Department
Contact: legal@eightfold.ai

Risk Assessment · Audit finding: Pass

How Eightfold AI identifies and monitors bias risk.

The audit reviewed the internal risk assessment process, including the risk register, the risk prioritization methodology, and evidence of ongoing monitoring. BABL AI reviewed risk register dashboards, meeting minutes, and testimony from the maintainers of the risk register for AI Interviewer. The risk assessment covered risk identification, stakeholder impact, severity, likelihood, risk sources, and controls.

Independent Auditor

Audited by an independent firm.

BABL AI Inc. is an independent AI auditing firm based in Iowa City, Iowa. Its lead auditors are ForHumanity Certified Auditors under the NYC AEDT Bias Audit standard. The BABL AI audit framework, the Criterion Audit Framework, is modeled after financial auditing practice and was published in the Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. BABL AI independence is codified by the Sarbanes-Oxley Act of 2002 and the ForHumanity Code of Ethics. Fees are unrelated to the opinion rendered.

Signed by BABL AI Inc., July 8, 2026.

Scope and limitations

What this audit covers, and what it does not.

This audit was conducted to satisfy the bias audit requirement of New York City Local Law No. 144 of 2021. The audit does not certify that the model is bias-free. No audit can make that claim. It is not intended to demonstrate compliance with any legislation other than the NYC AEDT law. The report does not make any determination as to whether the model is, in fact, an automated employment decision tool under the law.

Assessed in this audit

gender, race/ethnicity (seven groups), age, intersectional groups (all gender and race/ethnicity permutations), internal governance, and risk assessment.

NOT Assessed

protected classes beyond race/ethnicity, gender, and age, including immigration or citizenship status, disability status, marital and partnership status, national origin, pregnancy and lactation accommodations, religion or creed, sexual orientation, and veteran or active military service member status.

The race/ethnicity and gender intersectional results are based on synthetic test data, as described above.

This audit covers AI Interviewer. Our talent intelligence products are audited separately: see the Matching Model bias audit results. →

Our commitment

We do this every year.

New York City Local Law 144 requires an annual bias audit. Yes, we conduct these audits because the law requires it, but we also go beyond because responsible AI is an ongoing practice, not a one-time event.

Download full 2026 audit report (PDF) →