Opptly matches the right people
with right work, fairly

Passing an independent AI bias audit

By auditing its AI model with Pacific AI, Opptly can put independent fairness evidence in front of enterprise buyers and refresh it every quarter.

• Delivers independent audit evidence for enterprise security reviews and RFPs
• Tests against the strictest legislation governing hiring AI in its primary market
• Runs entirely on synthetic data inside Opptly’s own network
• Re-runs quarterly to catch score drift between annual audits

Delivering a regulatory grounded, rigorous, and independent fairness audit is what our clients deserve – and the standard we would like to see set for Generative & Agentic AI.

Jason Safley
Chief Technology Officer, Opptly

Bias audits exist to protect job seekers, employers, and AI vendors

If you build hiring AI, your customers inherit your testing gaps. If you deploy it, you answer for them.

EEOC Uniform Guidelines require adverse-impact analysis: selection rate and impact ratio calculations aligned with 29 CFR §1607.15, benchmarked against the four-fifths rule. Opptly’s audit tested 8 protected-attribute categories: sex and gender, race and ethnicity, national origin, religion, pregnancy and maternity status, age, disability, and military or veteran status.

NYC Local Law 144 requires an independent bias audit for automated employment decision tools, and specifies who may run one: not the team that built the tool.

The EU AI Act classifies employment screening tools like Opptly’s as high-risk under Article 6 and Annex III, then splits enforceable obligations across the supply chain rather than pinning them on one party.

Developers carry data governance obligations under Article 10, where bias detection in training, validation, and test data is mandatory, and produce the Annex IV technical file. Deployers commission the audit, publish the summary, and notify candidates. The two connect: the developer’s documentation feeds what the deployer needs to show.

Opptly built the matching engine itself, so no upstream vendor’s audit covers it. The model had to be tested directly.

An engine that has never been independently audited leaves a gap on both sides of that chain.

600+ regulations tracked and updated quarterly

Three separate vendors usually cover this ground, and their evidence does not reconcile when a regulator asks. Opptly runs it as one program.

Governor holds the registry, risk assessments, and policy management. The AI Policy Suite updates quarterly across 600+ regulations, standards, and frameworks at global scale, so obligations surface before an audit rather than after an inquiry.

Every metric traces to a named provision. Selection rate and impact ratio follow the published calculation guidance for automated employment decision tools. Adverse-impact analysis maps to EEOC 29 CFR §1607.15, and testing documentation to Article 10.3-5 and Annex IV.2(g).

Guardian runs continuous monitoring inside Opptly’s infrastructure, version-controlled and re-runnable by Opptly’s own team. Opptly holds the Pacific AI Unbiased Certification, valid 12 months while the tested model remains materially unchanged.

Independent LLM Audit: Methods, Statistics, and Design Choices

The question a regulator or an enterprise buyer will ask is not whether you tested. It is whether your test could have found bias if bias were there.

The design answers that directly. Take one resume, change exactly one thing about the candidate’s identity, and score it again. A name that signals gender or ethnicity. A graduation year that signals age. An assistive-technology reference that signals disability. Everything else stays identical, so nothing except that one signal can explain a change in score.

Run that 148,940 times across 43 protected classes and the result is unambiguous: no evidence of algorithmic bias under the evaluation criteria established by New York City’s Local Law 144, U.S. federal EEOC standards, and the European Union’s AI Act. Selection rates held between 0.91 and 0.92. Impact ratios landed near 1.00. Every class measured well inside the range where a difference is too small to affect who gets shortlisted.

One check shows the robustness of the method. A preliminary run on 10% of the data showed an apparent signal at high scores. At full scale it narrowed to the extreme tail and was attributed to tail noise. A smaller study could have published a false positive as a real finding.

Opptly is a pioneer in proving its own AI models are unbiased, and we’re proud to provide them with the statistically rigorous evaluation that measures this across regulations and subgroups.

David Talby
CEO, Pacific AI

About the companies

Opptly’s proprietary AI platform combines workforce-specific data, advanced machine learning, and deep skills intelligence to deliver accurate skills-based matching, workforce analytics, career progression insights, and workforce optimization. By leveraging purpose-built AI, transparent skills reasoning, and an intuitive user experience, Opptly helps organizations strengthen strategic workforce planning, improve role relevancy to the market, hire better talent, and stay competitive in a rapidly evolving labor market.

Pacific AI provides the AI governance platform and independent assurance services that organizations use to validate, monitor, and manage AI systems against the regulations that apply to them. Governor, Gatekeeper, Guardian, and the AI Policy Suite cover the full system lifecycle from policy tracking through pre-release testing to production monitoring.