This FAQ provides you — the customer acting as data controller — with technical and processing information about the Greenhouse Voice AI feature (formerly offered by Ezra AI Labs Inc. and now part of Greenhouse following the Greenhouse acquisition of Ezra AI Labs) to support your evaluation of the feature. This document also provides the technical foundation you need to conduct data processing assessments, including Data Protection Impact Assessments (DPIAs) under the GDPR and analogous assessments under other frameworks, as you determine may be appropriate.
Note: None of this is intended to be legal advice, and none of the answers provided are intended to override or contradict the advice of your legal counsel.
Voice AI is a feature governed by the Greenhouse MSA, the Greenhouse AI Terms, the Greenhouse DPA, the Greenhouse Security Addendum, and Voice AI Terms (Product Terms). In the event of any conflict between this FAQ and those documents, the executed contractual documents control.
Feature overview
What is Voice AI?
Voice AI is an optional Greenhouse feature that conducts video interviews with job applicants using an AI agent. It presents recruiter-configured questions to the applicant, records the audio and video of the applicant's responses, generates a transcript of those responses, and compares the transcript against rubric criteria pre-established by the recruiter to produce a recommended score (the "Voice AI Score") that recruiters and hiring managers may accept, modify, or disregard. The feature is designed as assistive AI — it does not automatically advance or reject candidates, and hiring decisions remain with recruiters and hiring managers. The audio and video recordings themselves are not used to produce the Voice AI Score.
What technology powers Voice AI?
Voice AI combines real-time speech-to-text, large language models (LLMs) for transcript analysis and rubric scoring, text-to-speech for the AI agent's spoken responses, and a video-conferencing layer for the interview session. The architecture uses an orchestration layer (Pipecat) that routes audio through specialized third-party AI model providers (for example, speech-to-text via Deepgram and LLM analysis via OpenAI and/or Google) and a WebRTC-based media transport (Daily.co) for the candidate-facing video session. See Third-party processing (subprocessors) below for the full subprocessor list and roles.
What are the excluded functions of Voice AI?
Voice AI is contractually scoped to exclude the following functions. Voice AI does not:
- Perform emotion recognition, sentiment inference, affect recognition, voice-based personality inference, or any analogous inference of an applicant's emotional or psychological state based on voice tonality, facial expression, gaze, pauses, or other non-verbal signals
- Extract, generate, store, or analyze biometric identifiers, and does not use captured audio or video to identify or verify a specific individual by means of biometric analysis (no voiceprint or faceprint extraction)
- Use audio, video, or facial data to infer protected characteristics
- Make, and is not designed to make, final hiring decisions
Analysis of applicant responses is conducted against the transcript derived from the captured audio; the underlying audio and video files are used only to generate that transcript and to provide the applicant interview record back to the recruiter.
Data processing details
What are the roles of the parties under data protection laws?
Greenhouse is a data processor (or service provider, as applicable) acting on behalf of its customers to process the personal data relating to the customer's job applicants and personnel. The customers are the data controllers (or businesses, as applicable), who decide why and how personal data is processed.
What data sources does Voice AI access?
The primary source of personal data is the candidate, who provides audio and video responses during the Voice AI session and may provide additional information (such as resume content) earlier in the application process. Additional sources include: the recruiter's ATS (whether Greenhouse Recruiting or a third-party ATS), which supplies candidate identifiers and job/interview metadata; and the recruiters and hiring managers, who configure the questions, rubrics, and "ideal candidate" profile that drive evaluation.
What specific personal data does Voice AI process?
In connection with the Voice AI feature, Greenhouse processes the following categories of applicant personal data, to the extent provided by the recruiter or captured from applicants:
- Identifiers and contact data: name, email address, phone number (if provided), role/title, company, user IDs, ATS candidate IDs and job IDs, recruiter user IDs
- Application content: employment history and resume content provided to the customer
- Interview scheduling data: interview invitations, links, and timestamps
- Interview media: audio and video recordings of interviews conducted through Voice AI
- Derived interview artifacts: transcripts generated from those recordings; Voice AI Scores generated by comparing the transcript against the customer-configured rubric; question-by-question scoring rationale; identified strengths and opportunities-for-growth statements; cheat- and fraud-detection signals (provided to the recruiter as decision-support only and not factored into the overall Voice AI Score)
- Account and usage data: authentication data, role/permissions, event logs (for example, sign-in, session state, error/crash logs), feature usage
- Device and technical data: IP address, device/browser/OS details, network diagnostics, security/fraud signals
- Billing data (customer contacts only, not candidates): subscription/invoice details, payment method tokens
- Support content: tickets and messages voluntarily provided by the customer
Does Voice AI process biometric (special category) data?
As described in What are the excluded functions of Voice AI? above, Voice AI does not derive voiceprints, faceprints, or other biometric identifiers from captured audio or video. Audio recordings of voice may, in certain jurisdictions and on certain analyses (for example, voiceprinting), be classified as biometric data; Voice AI does not perform such analyses.
Does the system process structured application data?
Voice AI's evaluation algorithm operates on the candidate's interview transcript, the customer-defined question rubric, and the customer-defined ideal-candidate profile. Resume content and structured application data may also be made available to the system to provide context for the interview but are not the basis of the Voice AI Score itself.
Data retention and storage
How long are interview recordings, transcripts, and scores retained?
Except as otherwise outlined in your contract or documentation, these are retained for the duration of the contract unless deleted earlier by you.
How does retention interact with the Illinois Artificial Intelligence Video Interview Act?
Under the Illinois Artificial Intelligence Video Interview Act, 820 ILCS 42/15(a), when a candidate requests destruction of their video interview, the employer must destroy — and request that vendors destroy — the recording within 30 days. Greenhouse will delete applicant audio, video, transcripts, and Voice AI Scores within 30 days of a customer-initiated deletion request (or of a candidate request relayed by the customer), and will instruct applicable third-party AI model providers to do the same.
What are the customer's retention obligations as data controller?
Customer action: You're responsible for configuring retention settings consistent with applicable law and your own legitimate business needs. Consider applicable employment-records retention laws, the Illinois AI Video Interview Act 30-day deletion-on-request rule, GDPR storage limitation, CCPA/CPRA disclosed-retention requirements, and any litigation-hold obligations.
Under Greenhouse's analysis, the retained transcript and scoring data satisfy hiring-related employment-records retention obligations; longer raw audio/video retention isn't required to meet those obligations and should be justified by a separate, documented purpose.
Third-party processing (subprocessors)
What data is shared with subprocessors?
Subprocessors receive only the data necessary to perform their function. Speech-to-text providers receive audio streams to produce transcripts. LLM providers receive transcript text (and, where applicable, the customer-configured rubric and ideal-candidate profile) to produce the Voice AI Score and scoring rationale. Text-to-speech providers receive the text the AI agent is to speak. Video-transport providers route real-time audio/video streams between the candidate and Greenhouse. Cloud-infrastructure subprocessors host data at rest. Fraud-and-verification subprocessors receive limited signals (for example, IP, phone number, email) for integrity-check purposes.
What protections exist for subprocessor processing?
Greenhouse maintains data processing agreements (DPAs) with each subprocessor that flow down all Greenhouse data protection obligations, prohibit subprocessors from using personal data for their own purposes (including to train or improve their own machine-learning models), and require adequate technical and organizational security measures.
Is customer data used to train Greenhouse or third-party AI models?
No. Greenhouse doesn't train on customer data related to this feature. Personal data isn't used to train Greenhouse's or any third party's models.
How does Greenhouse notify customers of changes to the subprocessor list?
Greenhouse provides at least 30 days' advance notice of the engagement of any new subprocessor for the Voice AI feature, via the Greenhouse Trust Portal subprocessor notification mechanism. Customers may object on reasonable, data-protection-related grounds.
International transfers
How are international data transfers handled?
Greenhouse is certified to the EU-U.S. Data Privacy Framework and the UK Extension to the EU-U.S. Data Privacy Framework. Cross-border transfers of personal data from the EEA or the UK are subject to those frameworks. Greenhouse secondarily relies on Standard Contractual Clauses and additional safeguards for data transfers.
Security measures
What encryption protects candidate data?
Voice AI data is encrypted in transit between the candidate, Greenhouse, and Greenhouse subprocessors using Transport Layer Security (TLS) version 1.2 or higher (TLS 1.3 in transport to voice-agent infrastructure). Data is encrypted at rest using a minimum Advanced Encryption Standard (AES) 256-bit encryption.
What technical and organizational measures protect candidate data?
Greenhouse implements technical and organizational measures as described in Annex II of the Greenhouse DPA. Greenhouse is also certified to ISO 27001, ISO 27701, ISO 42001, and SSAE 18 SOC 2 standards by independent third-party auditors. More information about Greenhouse's security program and practices is available at the Greenhouse Trust Portal.
What access controls are in place?
Access to Voice AI audio, video, transcripts, and scoring data is available to Greenhouse production engineers to enable Greenhouse's provision of its services. Access to this data is restricted with the same permissions required to access any other Greenhouse customer production data. Greenhouse technical support users don't have access to candidate data unless the customer grants access via a "Temporary Access Grant" mechanism and the support user logs in as a customer employee.
What AI-specific security measures are implemented?
Greenhouse follows generative-AI security best practices, including mitigations for prompt-injection attacks, output-handling vulnerabilities, denial-of-service against AI components, and data-leakage risks (including risks specific to RAG pipelines and tool-using agents). Guardrails are implemented at the prompt and system-design layer to limit Voice AI's context to the minimum data necessary to evaluate the candidate against the rubric. Adversarial test cases exist for prohibited questions (for example, questions about protected characteristics, current salary, or illegal interview topics) and for resilience to attempts to derail the interview off-rubric.
What vulnerability management process applies?
Greenhouse maintains a vulnerability management program covering Voice AI code and its third-party dependencies. Dependency scanning (Dependabot) and secret scanning are integrated into the CI/CD pipeline; static analysis (SAST) is applied to all pull requests. Vulnerabilities are tracked and remediated under documented SLAs, with critical and high-severity vulnerabilities prioritized for accelerated remediation.
Individual rights implementation
How can candidates exercise their data subject rights?
Customer action: You must establish processes for candidates to exercise their rights under applicable data protection laws, as advised by your legal counsel. This should include clear contact information and procedures for handling access, correction, deletion, and objection requests related to AI processing.
Can candidates request correction of their Voice AI results?
Customer action: As data controller, you determine how candidates contact you to request correction. As a matter of product functionality, your recruiters and hiring managers may override or adjust any Voice AI Score before a hiring decision is made (overriding or adjusting a score triggers a text box to explain the rationale for the change); the Voice AI Score is a recommendation, not a decision. Candidates may also request human review of a Voice AI Score, and Voice AI surfaces a mechanism for candidates to request such review within the post-interview email flow.
How are deletion requests handled?
Customer action: As data controller, you determine how candidates contact you to request deletion. As a matter of product functionality, when you initiate deletion of an applicant's data, Greenhouse deletes the audio, video, transcript, Voice AI Score, and related derived artifacts, and instructs applicable third-party AI model providers to do the same, within 30 days of the deletion request. This aligns with the 30-day deletion-on-request requirement of the Illinois AI Video Interview Act.
What opt-out mechanisms exist for Voice AI?
You can make Voice AI optional for candidates. Greenhouse provides candidate-facing email and notice templates that include an explicit opt-out instruction. A candidate who opts out is routed to a customer-defined alternative selection process and isn't penalized, disadvantaged, or subject to any adverse inference for the opt-out.
Customer action: You may be required to design and operate the alternative selection process and to ensure it's presented to candidates who decline to consent to Voice AI, who request an alternative based on a disability, religious observance, or other protected status, or who are otherwise entitled to an alternative under applicable law.
Automated decision-making
Does Voice AI make fully automated decisions about candidates?
No. The use of Voice AI isn't an automated decision under GDPR Article 22. Voice AI doesn't make, and isn't designed to make, any hiring decision. The Voice AI Score is a recommendation produced by comparing the candidate's transcript against the customer-configured rubric; the score is presented to the customer's recruiters and hiring managers, who may accept, modify, or disregard it. All decisions to advance, reject, rank, screen out, schedule, extend an offer to, or decline to extend an offer to any applicant are made by the customer's human personnel.
What human oversight mechanisms are in place?
The Voice AI Score, the transcript, and (where retained) the audio/video are surfaced to authorized customer personnel in the customer's Greenhouse account. Reviewers may override or modify any score (overriding or adjusting a score triggers a text box to explain the rationale for the change). Voice AI Terms expressly prohibit using the Voice AI Score as the sole basis for any employment-related decision.
Customer action: You must ensure that all decisions informed by a Voice AI Score are reviewed by qualified human personnel with authority to accept, modify, or disregard the score, and that those personnel have received sufficient training to understand Voice AI's design, limitations, and appropriate use.
How does Voice AI provide transparency about its scoring?
For each interview, Voice AI surfaces: an overall Voice AI Score (a weighted average of question-level scores), question-by-question scores against the customer-configured rubric, a written scoring rationale citing specific candidate responses, identified strengths and opportunities for growth, the full transcript with question timestamps, and (where retained) the underlying audio and video. Cheat- and fraud-detection signals are surfaced as supplemental, decision-support information only and aren't factored into the Voice AI Score.
Candidate notices and opt-out
What notice is provided to candidates before a Voice AI session?
Customer action: Under Voice AI Terms, you must provide clear, prominent, and timely notice to each applicant, before the start of the Voice AI session, that:
- The session will be conducted by an AI interviewer
- The applicant's audio and video will be recorded
- A transcript will be generated from the audio and analyzed by AI to produce a score
- A human will review the score and make the final decision
- The applicant may request an alternative interview process
- The applicant may request access to, correction of, or deletion of their data
Greenhouse provides candidate-facing notice and email templates. These templates are illustrative and provided as a starting point only — you're responsible for tailoring the notice to your jurisdiction and use case.
What recording consent must be captured?
Customer action: Depending on the jurisdiction in which your applicants are located, you may need to obtain affirmative, specific, and separately capturable consent from each applicant, before the start of any Voice AI session, to the recording of audio and video. For example, the Illinois AI Video Interview Act and all-party-consent recording statutes in certain U.S. states require such consent. You may also be required to maintain an auditable record of each consent obtained, including the form and timestamp of the consent, the version of the notice presented, and an identifier tying the consent to the applicant. Consult your legal counsel to determine what consent obligations apply in your jurisdictions.
Where is the in-product transparency notice presented?
Voice AI presents a transparency notice on the candidate's pre-interview landing page (before the candidate enters the interview room) and an in-session transparency banner during the interview itself.
Data breach notification
How does Greenhouse handle data breach notification?
Upon becoming aware of a personal data breach affecting customer personal data processed in connection with the Voice AI feature, Greenhouse will notify the affected customer(s) without undue delay, and in any case within 72 hours. The notification will include, to the extent then known and as it becomes known: the nature of the personal data breach; the categories and approximate number of data subjects affected; the categories and approximate volume of personal data affected; the likely consequences of the incident; and the measures taken or proposed to address the incident and mitigate possible adverse effects.
Who is responsible for notifying regulators and data subjects after a breach?
Customer action: As data controller, you're responsible for notifying the relevant supervisory authority and, where required, affected data subjects of a personal data breach under Articles 33 and 34 of the GDPR (and analogous laws). Greenhouse will provide commercially reasonable assistance in meeting those notification requirements, including by providing the information described above.
Bias mitigation and fairness
What bias-mitigation measures are implemented in Voice AI?
Greenhouse implements the following bias-mitigation measures for Voice AI:
- Voice AI is restricted by design from asking questions about protected characteristics or other prohibited topics (including current-salary questions in jurisdictions that prohibit them); adversarial test cases verify this restriction
- Scoring is performed against a customer-configured, role-specific rubric and the candidate's transcript, not against demographic signals, voice characteristics, or facial features
- Cheat- and fraud-detection outputs are surfaced as decision-support only and aren't factored into the Voice AI Score
- Customers may override or modify any Voice AI Score before a hiring decision is made
How frequently are bias audits conducted?
The bias audit is conducted every time Greenhouse releases a change to the Voice AI algorithm. Greenhouse also uses a third-party vendor, WardenAI, to conduct monthly bias audits.
Are bias audit results available?
Yes. The third-party bias audit results are publicly available at trust.warden-ai.com/greenhouse/ai-interviewer.
AI hiring law compliance
What AI hiring laws apply to Voice AI?
Voice AI may trigger obligations under, among other laws:
- Illinois Artificial Intelligence Video Interview Act, 820 ILCS 42 (notice, explanation of how the AI works, consent, sharing restrictions, 30-day deletion on request)
- Illinois Human Rights Act as amended by HB 3773 (in force January 1, 2026)
- New York City Local Law 144 of 2021 (annual bias audit, candidate notice, alternative process disclosure)
- California ADMT regulations under the CCPA (notice, opt-out, ADMT access rights)
- EU AI Act (Voice AI is likely classified as a High-Risk AI System under Annex III, with related conformity-assessment, transparency, logging, human-oversight, and registration obligations)
- Ontario Working for Workers Four Act (candidate notice)
Customer action: As the data controller, you're solely responsible for ensuring that your use of Voice AI complies with all applicable AI hiring laws (as defined in the Voice AI Terms). Greenhouse provides product capabilities (notice templates, opt-out flows, human-override controls, deletion-on-request workflows, audit-log surfaces) and this FAQ to support your use of the feature.
What is Voice AI's status under the EU AI Act?
Greenhouse anticipates that Voice AI will be classified as a High-Risk AI System within the meaning of Annex III of the EU AI Act when used to evaluate candidates for employment. Greenhouse's posture toward the EU AI Act's high-risk obligations — risk-management system, data governance, technical documentation, record-keeping, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, post-market monitoring, registration in the EU database, and conformity assessment — is the subject of an active readiness program.
Note: The high-risk-AI-system enforcement date for Annex III systems has been deferred to December 2, 2027, subject to publication in the Official Journal.
Legal basis and compliance
What legal basis should I use for processing through Voice AI?
Customer action: As the data controller, you should determine your lawful basis for processing (for example, explicit consent, contract, or legitimate interest) under applicable data protection law in coordination with your legal counsel.
Do I need to conduct a Data Protection Impact Assessment (DPIA)?
Customer action: As the data controller, you should make this determination in coordination with your legal counsel. Using an AI-evaluated video interview tool to evaluate candidates may require a DPIA under Article 35 of the GDPR (and equivalent assessments under UK GDPR or other frameworks), and/or other assessments required under laws such as data protection law, AI law, or employment law. Greenhouse provides this FAQ to give you reasonable assistance in completing the assessments you deem necessary.
What notice must I provide to candidates about the use of Voice AI?
Customer action: As the data controller, you should make this determination in coordination with your legal counsel. You may need to provide clear notice to candidates about AI processing before they submit applications, as advised by your legal counsel and as required by applicable laws. This generally includes the purpose of AI processing, candidates' rights, and how to exercise them, and may include: that the session is conducted by an AI interviewer; that audio and video will be recorded; that a transcript will be generated and analyzed; that a human reviews and makes the final decision; and that the candidate may exercise data subject rights. See Candidate notices and opt-out above.
Greenhouse provides templates as a starting point; you're responsible for ensuring your notice complies with applicable law in the relevant jurisdiction. Consider implementing job board disclosures and updating your privacy policy to address AI use in recruitment.
Greenhouse provides a settings page that lets you configure a custom AI disclaimer for job posts.
Risk assessment support
What risks should I document in my DPIA or impact assessment?
Customer action: As the data controller, you should make this determination in coordination with your legal counsel. Risks to consider may include:
- Accuracy of Voice AI scoring for your specific candidate population
- Potential bias in your specific use case
- Candidate awareness, consent, and ability to elect an alternative process
- Data security in transit and at rest
- Compliance with consent requirements in wiretap statutes for candidates in two-party-consent U.S. states, as applicable
- Compliance with candidate rights, such as rights to data deletion and associated timelines
- Compliance with local employment laws
- Impact on candidate experience
Assess likelihood and severity based on your specific context.
What safeguards should I implement?
Customer action: As the data controller, you should make this determination in coordination with your legal counsel. Based on your risk assessment, consider implementing:
- Clear pre-interview notice and recording consent processes
- A non-AI alternative selection process for candidates who decline Voice AI or request an accommodation
- Training for recruiters and hiring managers on Voice AI's design, limitations, and appropriate use
- Documented bias-monitoring procedures
- Retention configuration aligned with local laws and your legal obligations
Monitoring and governance
What ongoing monitoring should I implement?
Customer action: You should make this determination in coordination with your legal counsel. Establish processes to monitor: recruiter compliance with human-oversight requirements; candidate feedback and complaints; Voice AI scoring accuracy for your specific roles; any patterns suggesting disparate impact in your hiring outcomes; and compliance with your retention settings.
How often should I review my assessments?
Customer action: You should make this determination in coordination with your legal counsel. You may want to schedule regular assessment reviews based on: a change in your use of Voice AI; an update to applicable AI hiring laws; a published Greenhouse change to Voice AI that may materially affect the assessment; feedback or complaints from candidates or regulators; or the engagement of a new subprocessor.
Documentation and audit support
What records does Greenhouse maintain?
Greenhouse retains rubric configurations, system logs, scoring rationale, and human-override events sufficient to support transparency, troubleshooting, and bias-audit and governance processes, in accordance with Greenhouse's internal retention policies. These configurations and logs relate to job-level criteria and system behavior and are governed separately from candidate-record retention.
What documentation should I maintain?
Customer action: You should determine your documentation set in coordination with legal counsel. Typical records include: your DPIA and any other impact assessment; legal-basis determination and (where applicable) the legitimate-interests balancing test; candidate notices and consent records; staff training records; bias-monitoring and audit results; candidate rights requests and responses; and records of retention configuration and deletion activities.
Additional technical information for customer reference
Score generation and human override
Voice AI's overall score is a weighted average of question-level scores (each 1–4), with weights set by the customer's importance rating (1–5) on each question. Each question score is produced by comparing the candidate's transcribed response to the customer-configured rubric (Great / Good / OK / Poor answer descriptions) and to the customer's "ideal candidate" description. Recruiters may override any score in the Greenhouse account; overrides are logged in-product.
Cheat and fraud detection
Voice AI surfaces, as decision-support only, signals related to (a) integrity of interview responses (for example, patterns consistent with reading or scripting) and (b) limited metadata-based authenticity signals (for example, IP, email, phone). These signals aren't factored into the Voice AI Score and are presented to recruiters with clear labeling. Customers should ensure that recruiters don't use cheat- or fraud-detection signals as the sole basis for any adverse action.
Subprocessor notifications
Customers can subscribe to subprocessor change notifications via the Greenhouse Trust Portal. Greenhouse provides 30 days' advance notice of any new subprocessors.