> This is the markdown version of https://www.classet.ai/blog/legal-teams-ai-screening > Learn more at https://www.classet.ai ![How legal and compliance teams evaluate Classet's AI screening](/_next/image?url=%2Fimages%2Fblog%2Flegal-teams-ai-screening.png&w=3840&q=75) # Why Legal Teams Approve Classet's AI Screening The document to hand your legal, security, and compliance reviewers: how Classet uses AI to speed up hiring operations without ever making a hiring decision. [![Paul Jones](/_next/image?url=https%3A%2F%2Fassets.basehub.com%2Fe0b5701f%2F6599306507912123f90f150a8bfaaf6c%2Fscreenshot-2026-01-28-at-10.53.16-am.png%3Fwidth%3D100%26height%3D100%26quality%3D100&w=96&q=75) Paul JonesHead of Growth at Classet ](/blog/authors/paul-jones) August 4, 2026 AI Recruiting, Guides & Insights Your talent team wants to move forward. Legal wants to understand what role the AI actually plays before anyone signs. That review is the right instinct, and this post is written for the people running it: general counsel, security, and compliance. It lays out exactly what Classet's AI does, what it is built not to do, and the documentation that backs both claims up. **TLDR for Legal, Security & Compliance:** - Classet's AI interviewer (Joy) never makes a hiring decision. It conducts structured phone interviews and organizes responses. A human recruiter makes every advance-or-reject call. - The only automated routing is employer-defined, objective knockout criteria: work authorization, license validity, shift availability. No scoring of tone, accent, or personality. - An independent third party (Warden AI) continuously monitors Classet for disparate impact, with results published on a public assurance dashboard rather than a once-a-year snapshot. - Voice screening captures conversation content, not biometric templates, which keeps it outside biometric statutes like Illinois BIPA and Texas CUBI. - Classet is SOC 2 Type II certified, and every interview produces a time-stamped recording, transcript, and summary synced to your ATS as the system of record. ## The Question That Decides the Review Every AI hiring law on the books turns on one distinction: does the tool make or substantially influence the employment decision, or does it automate the work around that decision? NYC Local Law 144 regulates "automated employment decision tools." The EU AI Act requires human oversight of "consequential decisions." Colorado SB 24-205 triggers when AI plays a "consequential role." The pattern is consistent: regulators care most about decision-making authority, because that's where discrimination claims live. So the first question to ask any AI screening vendor is not "how accurate is the AI?" It's "who decides?" At Classet, the answer is structural, not a policy promise. [Joy](/product/joy-ai-powered-ats), our AI interviewer, has no mechanism to rank candidates against each other, score them on subjective traits, or reject anyone based on its own judgment. The system speeds up the operations of hiring: answering the phone in seconds instead of days, asking every candidate the same structured questions, transcribing and summarizing responses, and writing everything back to your ATS. The decision itself stays with a person on your team. ## What Joy Does, and What It's Built Not to Do The clean way to review this is as two lists. **Joy does:** - Conduct structured voice interviews, asking every candidate for a role the same questions in the same order - Answer candidate questions about the role using content your team approved - Apply knockout criteria your team defines: binary, objective requirements like work authorization, a valid driver's license, or availability for a night shift - Produce a recording, transcript, and structured summary of each interview - Route candidates based on knockout results and surface everything else for human review **Joy never:** - Ranks candidates against each other - Scores tone, accent, personality, enthusiasm, or "culture fit" - Analyzes facial expressions or video of any kind - Rejects a candidate on its own judgment - Infers demographic traits about an individual, or uses them in any candidate-level outcome Knockouts deserve a closer look because they're the one place automation touches routing. A knockout is a requirement your team would enforce anyway: you cannot hire a driver without a CDL, so a candidate who says they don't have one is routed out, with the reason logged. That's the same screen a recruiter applies manually, made consistent. Because the criteria are binary and job-related, they're also the easiest part of the system to defend: there's no model weighing subtle signals, just a documented requirement and a documented answer. ## How the Bias Reduction Actually Works Most bias in early-stage screening doesn't come from malice. It comes from inconsistency and proxy signals. One candidate gets a rushed five-minute call, another gets twenty relaxed minutes. Resumes carry names, zip codes, and school names that [correlate with race and class](/blog/hear-them-out-ai-phone-screenings-edge-in-reducing-hiring-bias) even when nobody intends to use them that way. Structured phone screening removes several of those channels at once. Every candidate gets the same questions, the same patience, and the same interview at 9 PM from a truck cab as at 10 AM from a desk. Joy doesn't see a resume header, a photo, or a neighborhood. What reaches your recruiter is what the candidate actually said about the actual requirements, in their own words, with the recording attached. We're careful about the claim here: consistency reduces the surface area for bias; it doesn't prove absence of bias. Proof requires measurement by someone with no stake in the answer. Which is why the audit matters more than the architecture. ## The Third-Party Bias Audit, Spelled Out Classet partners with [Warden AI](https://trust.warden-ai.com/classet/ai-phone-interviewer), an independent AI assurance provider, for continuous bias monitoring. Here's how it works in practice: **What's measured.** The audit tests for disparate impact: whether outcomes differ across protected groups like race and gender, even when the process looks neutral on its face. This is the same standard the EEOC applies under Title VII, usually expressed through the four-fifths rule on selection rates. **How demographics are estimated.** Auditors need demographic distributions to test for disparate impact, but voluntary EEOC self-identification surveys have notoriously low completion rates, which leaves audits built on them with thin data. So the audit uses inference-based proxy methods, including name-based inference and Bayesian Improved Surname Geocoding (BISG), to estimate demographics in aggregate across the full applicant population. Regulators including the CFPB have long recognized these methods for exactly this monitoring purpose. **The line that matters for your review:** these inferred demographics exist only inside the audit, at the aggregate level. They are never attached to an individual candidate and never touch any screening outcome. Joy has no access to them. **Cadence and visibility.** Monitoring is ongoing, not an annual event, and results are published on Warden AI's public assurance dashboard. Your team can check the current results before a demo, during procurement, and after deployment, without asking us for a report. A once-a-year audit tells you a tool was fair last spring; continuous monitoring catches drift between snapshots, which is what Colorado's impact-assessment regime and the EU AI Act's output-monitoring rules are pushing the industry toward anyway. ## Phone, Not Video: The Medium Lowers Exposure Some AI screening risk comes from the modality, not the model. Video interview tools analyze facial data, which pulls them into biometric privacy statutes like Illinois BIPA and Texas CUBI, laws with private rights of action and statutory damages. Voice conversations that capture content rather than biometric identification templates sit outside those statutes. The medium also produces a better audit trail. Every Classet interview generates a linear, time-stamped record: recording, transcript, structured summary. Your ATS remains the system of record, so those artifacts land in the same compliance trail you already maintain, rather than in a vendor silo. If a decision is ever challenged, you can show exactly what was asked, what was answered, and which human made the call. ## The Security Review For the security half of the evaluation: Classet maintains SOC 2 Type II compliance, meaning an independent auditor has evaluated the operating effectiveness of our controls over time, not just their design at a point in time. Access to customer data is role-based and permissioned, audit logs cover security-relevant events, and we operate a defined incident response and breach notification process. Compliance materials and supporting documentation are available through our Trust Center for teams running a formal review. ## What We Bring to Your Review We'd rather you run a hard review than a fast one, so we don't wait to be asked. Every legal and security review gets the same package: 1. Current Warden AI assurance results, including the disparate impact methodology 2. SOC 2 Type II report and Trust Center access 3. A walkthrough of knockout configuration, so your team can verify criteria are binary and job-related 4. Sample interview artifacts: recording, transcript, and summary as they appear in your ATS 5. Data retention and deletion terms 6. Our position on candidate notice requirements in the jurisdictions where you hire That's the package any vendor in this category should be able to produce without friction, and if one can't, that tells you something. If you want the fuller version of this exercise, we've published the [questions to ask any AI recruiting vendor](/blog/ai-recruiting-vendor-questions) and a breakdown of [what the 2026 wave of AI hiring laws requires](/blog/ai-recruiting-bias-audits-compliant-platform), including NYC Local Law 144, Colorado SB 24-205, Illinois HB 3773, and the EU AI Act. The short version: laws differ on notices and audits, but all of them get easier when the AI's role is operational and the decision is documented as human. ## FAQ Does Classet's AI ever reject a candidate? Joy routes candidates based on employer-defined knockout criteria: binary, objective requirements like work authorization or license validity that your team configures and would enforce in any process. Beyond knockouts, Joy has no rejection mechanism. Every judgment-based advance-or-reject decision is made by a human recruiter reviewing the candidate's actual answers. Is Classet an automated employment decision tool under NYC Local Law 144? Classet is designed to meet or exceed AEDT standards including NYC Local Law 144, and is continuously monitored by an independent third party for disparate impact. Because classification questions depend on how you deploy any tool, we'll walk your counsel through our configuration and documentation so you can make that determination with full information rather than a vendor's assertion. How can you audit for bias without asking candidates their race or gender? The third-party audit uses inference methods accepted by regulators, including name-based inference and Bayesian Improved Surname Geocoding (BISG), to estimate demographic distributions in aggregate. This avoids relying on voluntary self-identification surveys with low response rates. Inferred demographics are used only for aggregate fairness monitoring, never attached to an individual or used in any screening outcome. Does voice screening trigger biometric privacy laws like BIPA? Classet interviews capture conversational content, not biometric identifiers like voiceprints used for identification or facial geometry. That keeps voice screening outside the scope of Illinois BIPA, Texas CUBI, and similar biometric statutes that video-analysis tools have to comply with. What records exist if a hiring decision is challenged? Every interview produces a time-stamped recording, full transcript, and structured summary, synced to your ATS as the system of record. Combined with logged knockout criteria and the human decision in your ATS, you can reconstruct exactly what was asked, what was answered, and who decided. Is Classet SOC 2 certified? Yes. Classet maintains SOC 2 Type II compliance, verified by an independent auditor over time rather than at a single point. Security documentation is available through our Trust Center for formal reviews. ## Key Points - The load-bearing question in any AI hiring review is who makes the decision. At Classet, it's always a human; the AI speeds up operations around that decision. - Automated routing is limited to binary, employer-defined knockouts. No ranking, no tone or personality scoring, no video analysis. - Warden AI independently and continuously monitors for disparate impact, using regulator-recognized inference methods, with results on a public dashboard. - Voice screening avoids the biometric statutes that apply to video tools, and produces a cleaner, ATS-native audit trail. - SOC 2 Type II certification and Trust Center documentation cover the security half of the review. ## Next Steps If your legal team has questions this post doesn't answer, bring them to the call. [Book a demo](/demo) and include your counsel or compliance lead: we'll walk through knockout configuration, the Warden AI dashboard, and the Trust Center live, so the review runs on evidence instead of vendor claims. [![Paul Jones](/_next/image?url=https%3A%2F%2Fassets.basehub.com%2Fe0b5701f%2F6599306507912123f90f150a8bfaaf6c%2Fscreenshot-2026-01-28-at-10.53.16-am.png%3Fwidth%3D100%26height%3D100%26quality%3D100&w=128&q=75) Paul Jones Head of Growth at Classet Paul comes from an operator background running an Alpine-owned company, and brings firsthand experience with the hiring challenges Classet was built to solve. He's driven by a belief that the right technology can make meaningful work more accessible. ](/blog/authors/paul-jones) Follow Classet in Google Add classet.ai as a preferred source and our hiring research is more likely to show up in your Top Stories and AI Overviews. [Add as preferred source →](https://www.google.com/preferences/source?q=classet.ai) ## Explore More ### Use Cases - [RPO / BPO Recruiting](/use-cases/call-centers-bpo) - [Healthcare Recruiting](/use-cases/healthcare) - [Hospitality Recruiting](/use-cases/hospitality) ### Integrations - [Greenhouse](/integrations/greenhouse) - [Bullhorn](/integrations/bullhorn) - [Lever](/integrations/lever)