Wellfound

Company overview

Wellfound is the startup-focused hiring marketplace formerly known as AngelList Talent. The talent business launched inside AngelList around 2012 and spun out as an independent company under the Wellfound brand in November 2022, having connected roughly 150k startups and 8M+ candidates at the time of the split. In April 2024 Wellfound acquired AI recruiting vendor Celential.ai and folded its talent graph into what is now the wellfound:ai product line: wellfound:ai Search / Reach (AI sourcing agents and outreach), wellfound:ai Applicants (AI resume screening that auto-generates 'must-have' and 'nice-to-have' criteria from a job description, ranks applicants and syncs decisions back to the customer's ATS), and Autopilot (human recruiters working on top of the same AI stack at $500/month per open role plus a 10% placement fee). A free built-in ATS anchors the bottom of the funnel; Reach seats run $135-$199/seat/month with a custom-priced 'Scale' tier, placing Wellfound firmly in the mid price band rather than enterprise. A video-based AI Interviews product is in beta on the candidate side. HQ is San Francisco, US.

Regulatory exposure

Wellfound's AI Resume Screening is squarely an automated employment decision tool: it produces an overall match score used to rank and filter applicants for employers, which brings it inside NYC Local Law 144 and, as an employment-decision AI, into Annex III high-risk territory under the EU AI Act, plus Illinois HB 3773 and Colorado's AI Act as those bite. Wellfound handles this unusually well on the audit axis and poorly on the documentation axis. On the plus side, it is a genuine Warden AI customer -- the continuous assurance dashboard at trust.warden-ai.com/wellfound/ai-resume-screening is linked directly from Wellfound's own site footer as 'AI audit center' on both reach.wellfound.com and wellfound.com/recruit/overview, it appears in Warden's Assured directory (listed 16 March 2026), it is audited monthly (latest 1 July 2026) across 23,584 candidate profiles for sex, race/ethnicity and intersectional bias using both disparate-impact and counterfactual-consistency methods, and a dedicated NYC LL 144 report page publishes impact ratios (0.96-1.00 across all groups). The material caveat is that Warden used its own independent candidate dataset rather than Wellfound's historical data 'due to a lack of access to historical data', and the audit covers only AI Resume Screening -- not the sourcing/ranking side, and not the AI Interviews beta, which is a video product and therefore also implicates the Illinois AI Video Interview Act and BIPA-adjacent consent questions. Against that, Wellfound publishes no AI policy, model card, explainability statement or responsible-AI page of its own; wellfound.com/recruit/all-features/wellfound-ai-overview contains no mention of bias, fairness, audits or human review at all. There is no EU AI Act statement, no Article 27 FRIA material, no deployer/employer compliance guidance, and no published AEDT notice tooling for the NYC customers who must post their own audit summaries and give candidates 10 business days' notice.

Path to a higher score

The audit work is already done and is a real asset -- what Wellfound is missing is first-party documentation to sit alongside it. Three moves would move the score materially. First, publish an Article 11-style system card on wellfound.com covering AI Resume Screening: intended purpose, inputs the model does and does not use, whether protected or proxy attributes are excluded, how match scores are computed, known limitations, and an explicit provider/deployer role split under the EU AI Act -- then link it from the wellfound:ai product pages rather than leaving those pages silent on fairness. Second, publish deployer guidance for employers: an NYC LL 144 pack (how to use Wellfound's Warden report, an AEDT candidate-notice template, the 10-business-day timeline), an EU AI Act Article 26/27 note with a FRIA starter template, and a public DPA -- this is the single largest gap and the cheapest to close. Third, extend the Warden scope beyond resume screening to the sourcing/ranking surface and to AI Interviews before that beta ships broadly, and move the audit onto Wellfound's own historical outcome data rather than Warden's synthetic-population proxy, which would remove the main methodological objection to the current LL 144 report. Smaller wins: surface per-criterion match reasoning and an audit log in-product, un-gate the trust.wellfound.ai contents (today the page renders nothing without JavaScript, so a prospective buyer cannot see the SOC 2 scope, subprocessors or retention terms), add an AI model-update changelog alongside the existing Atlassian status page, and state a data-exclusion and no-customer-data-for-training position publicly.