What it is
Warden AI is a third-party AI assurance company that tests HR-tech AI for bias, accuracy and explainability, then publishes the results on live dashboards instead of in an annual PDF. It was founded in 2023 by Jeffrey Pole and Eduard Schikurski, runs from London and Austin, and raised a £1.1M seed round in July 2025 led by the founders of Onfido, with Playfair participating. It is not an AI governance suite for every model in the enterprise, and it is not a law firm: it measures how a hiring model treats demographic groups and hands you evidence your counsel can use. Do not confuse it with the Warden Protocol crypto token, which has its own unrelated 2026 merger news.
The buyer question it answers: how do I prove our AI screener or AI interviewer isn’t discriminating under NYC Local Law 144, the EU AI Act and the state laws that follow?
How it works
- Dual-method bias testing. Each audit runs disparate impact analysis (do groups get systematically different outcomes?) and counterfactual testing (does swapping a name, gendered wording or a hobby change the score?). Greenhouse’s public bias audit statement describes exactly these two methods for its Talent Matching feature, tested across 10 protected classes on Warden’s own dataset.
- Its own test data. The Warden Dataset is consent-sourced HR data augmented with synthetic variations, so a vendor can be audited before it has a large customer base. Warden’s materials cite coverage of up to 14 or “15+” protected classes, including sex, race, age and disability.
- Continuous, not annual. PageUp’s June 2026 announcement describes monthly audits; Greenhouse commits that each new Talent Matching version passes an audit before release. Results sit at
trust.warden-ai.com/<vendor>, and vendors that keep passing appear in the Warden Assured Directory with a trust mark. - Pre-market testing in CI/CD. HTTP and file-based APIs let a vendor run checks in its release pipeline and show the result to buyers and regulators before launch.
Named vendor customers include Greenhouse, Beamery, Sense (verified January 2026), PageUp, Findem, Juicebox and Jack & Jill. Warden holds SOC 2 Type 2 and ISO 27001.
Pricing reality
Warden publishes no prices. Quotes are scoped by the number of AI systems, protected classes and regulations in scope, and a demo comes first. The useful benchmark is the market for a single NYC LL 144 bias audit: vendor-published price references run roughly $5,000–$50,000 per tool, with one specialist auditor listing a $10,000 starting fee per system and a six-week turnaround. That is a once-a-year deliverable. A monthly program with a public dashboard is a recurring subscription, so treat the one-off annual audit price as the floor you negotiate against, not the expected total.
For employers the cheaper route is often free: if your ATS or screening vendor already pays Warden, the dashboard for that feature costs you nothing to read.
Best for
HR-tech product and compliance leads selling AI screening, matching or interviewing into enterprise and New York City buyers, who need a public, repeatable answer to the security-questionnaire question “show me your bias audit”. Second fit: TA and legal leads at employers and staffing firms running several AI hiring tools who want one auditor across all of them.
Do not buy it if you need governance across non-HR models (credit, pricing, customer service); Warden is HR-only.
Versus the alternatives
Holistic AI is the broad AI governance platform with a dedicated NYC bias audit workflow. Pick it when HR is one of many AI risk domains and the buyer is a central AI governance office.
BABL AI and ORCAA are audit firms rather than software platforms. BABL ran the 2026 LL 144 audit of Eightfold’s matching model, and ORCAA can infer demographics when applicants never disclosed them. Pick one when you need a single signed annual audit and have no use for a live dashboard.
FairNow, acquired by AuditBoard in October 2025, is the route when AI risk should sit inside your existing GRC and audit tooling. Pick it if AuditBoard already runs your controls.
Pick Warden when the auditee is an HR-tech product, the audience is buyers, and continuous public evidence beats a yearly report.
Watch-outs
- A vendor’s dashboard is not automatically your LL 144 audit. Under the NYC rules, an employer can rely on a bias audit built on other employers’ data only if it supplied its own historical data or has never used the tool. Guard: ask the vendor whether your data is in the audited set; if not, commission an employer-level audit before you use the tool on New York City roles, repeat it within a year, and publish the summary as the law requires.
- The audit covers the feature the vendor chose. Greenhouse’s statement scopes Talent Matching only. Guard: list every AI feature you will switch on, and map each one to a named dashboard before go-live. Treat an unaudited feature as unaudited.
- Regulatory deadlines moved under the sales pitch. Colorado’s SB 26-189, signed 14 May 2026, replaced the old Colorado AI Act with a notice, human review and three-year record-keeping framework effective 1 January 2027 and dropped the mandatory impact assessment. The EU’s Digital Omnibus pushed Annex III employment obligations from 2 August 2026 to 2 December 2027. Guard: buy for NYC LL 144 and buyer due diligence now, and ask counsel which audits each jurisdiction actually requires before you pay for coverage you do not need.
- Warden is a seed-stage company. Guard: require in the contract that audit reports, dataset versions and audit-trail records are exportable, so your evidence survives a change of auditor.