Whitepaper
GDPR Compliance Guide for AI Interviews
Legal bases, applicants' rights, data retention, transfers and checklists for AI interviews.
Overview
This guide covers the data protection requirements for AI interviews in Europe. It is aimed at legal counsel, HR leaders and compliance officers and explains the legal bases, applicants' rights and the measures needed in day-to-day operation. It is based on regulatory guidance, case law and discussions with data protection authorities in several jurisdictions.
Contents
- 1 Summary: GDPR and AI recruiting
- 2 Chapter 1: Legal foundation – GDPR principles for recruiting
- 3 Chapter 2: Legal basis for processing candidate data
- 4 Chapter 3: Transparency and candidate rights
- 5 Chapter 4: Requirements for automated decision-making
- 6 Chapter 5: Data minimisation and retention
- 7 Chapter 6: International data transfers
- 8 Chapter 7: Implementation checklist and best practices
Summary: GDPR and AI recruiting
The European Union's General Data Protection Regulation (GDPR) is the most comprehensive data protection framework in the world and has a considerable impact on AI-supported recruiting. Companies that assess candidates with AI interviews have to meet complex requirements on data processing, transparency, automated decision-making and individual rights. Breaches have serious consequences: fines of up to €20 million or 4% of worldwide annual turnover (whichever is higher), reputational damage, litigation risk and a possible ban on processing candidate data, which would effectively bring recruiting to a standstill.
This white paper provides a comprehensive compliance framework developed together with data protection authorities, employment lawyers and privacy experts from several EU jurisdictions. The GDPR sets principles that apply throughout the EU, but the detail of implementation differs from country to country: Germany's additional safeguards for employee data, the French CNIL's guidance on AI recruiting and the precedents set by the Irish Data Protection Commission create differences that organisations need to know. The guide takes these differences into account and concentrates on practices that are compliant in every EU jurisdiction.
How personal data is handled also influences whether applicants trust the selection process. Organisations that respect privacy, are transparent about how they use data and commit to fair assessment win candidates' trust and strengthen their employer brand. Unclear or careless data processing can unsettle applicants and damage the company's reputation, even where individual formal requirements are met. The recommendations that follow aim at both: legal compliance and candidates' confidence in how their information is handled.
“GDPR breaches in recruiting can lead to fines of up to €20 million or 4% of worldwide turnover. That makes compliance a critical business priority.”
Chapter 1: Legal foundation – GDPR principles for recruiting
The GDPR lays down seven core principles that apply to all processing of personal data, including recruiting. Understanding them is the foundation for introducing AI interviews in a compliant way. This chapter applies each principle to recruiting.
Lawfulness, fairness and transparency
The first principle requires that data processing has a legal basis, is carried out fairly and is transparent to the people concerned. In recruiting, lawfulness typically rests on legitimate interests (the employer's interest in assessing candidates) or contractual necessity (the processing is needed to assess the application for employment). Fairness means that the processing does not disadvantage candidates unexpectedly and does not use data in ways they could not reasonably anticipate. Transparency requires clear information on which data is collected, how it is used and who has access.
For AI interviews, transparency means in concrete terms: candidates are told that an AI will assess their answers, which factors the AI assesses and whether a human reviews the result before decisions are made. It is not enough to hide the use of AI in long privacy policies – candidates must receive a prominent, clear notice with the interview invitation. Fairness is affected when AI systems apply different standards to different candidates or assess irrelevant factors such as accent or speech patterns instead of substance. Organisations must design AI interviews so that they concentrate on job-relevant competencies and assess all candidates consistently.
Purpose limitation
Purpose limitation requires that data is collected for specified, explicit and legitimate purposes and is not further processed in a way that is incompatible with those purposes. In recruiting this means: candidate data is collected to assess the application for specific positions and must not be repurposed for other activities. Using interview answers to train AI systems for unrelated purposes, sharing data with third parties other than the service providers needed for recruiting, or keeping data indefinitely for speculative future use may each breach purpose limitation.
In practice this means: privacy notices must list the uses of the data specifically, and organisations must limit actual processing to the stated purposes. Candidate data collected for one role must not automatically be considered for other positions without explicit consent or a clear notice that the organisation maintains a candidate database for future vacancies. AI training with candidate data calls for particular scrutiny: if training AI systems was not stated as a purpose when the data was collected, it is likely to be unlawful further processing. Organisations should either obtain separate consent for AI training purposes or state this use clearly in the initial notices.
Data minimisation
Data minimisation requires collecting only data that is adequate, relevant and limited to what is necessary for the stated purposes. In recruiting, this principle prohibits asking for information beyond what is needed to assess suitability for the job. Common breaches: requesting demographic information that plays no part in the selection decision, collecting excessive personal details or carrying out more extensive assessments than the job requirements justify. AI interviews are particularly sensitive here, because comprehensive conversation analysis can extract far more information than the questions explicitly ask for – sentiment, personality traits, communication patterns, stress indicators.
Compliance requires a careful analysis of which information is genuinely necessary for hiring decisions. Analyses or data points that are not used in assessing candidates should not be collected, or should be discarded immediately. If, for example, an AI interview system can analyse candidates' emotions but emotions are not relevant to the role, that capability should be switched off. Organisations should document the job relevance of all data collected and be able to justify its necessity when asked. If in doubt, collect less data – additional information can be gathered later in the process if it is really needed.
Accuracy, storage limitation, integrity and accountability
The remaining principles concern data quality, retention, security and compliance documentation. Accuracy requires reasonable steps to ensure that candidate data is correct and updated where necessary – especially important for AI systems, which can carry errors forward across several assessments. Storage limitation requires deleting candidate data once the recruiting purposes have been fulfilled, typically within 6 to 12 months for unsuccessful candidates, unless a specific business need and a legal basis support longer retention.
Integrity and confidentiality require appropriate security measures against unauthorised access, accidental loss and data breaches. Candidate data must be encrypted in transit and at rest, access must be restricted to authorised staff, and systems must be checked regularly for vulnerabilities. Finally, accountability requires organisations to demonstrate compliance through policies, training, impact assessments and documentation. This makes compliance an active duty: organisations must show that they have implemented appropriate measures and cannot simply claim compliance once they are challenged.
- Lawfulness: establish a legitimate basis for processing, typically legitimate interest or contractual necessity
- Transparency: give clear, prominent notice of the use of AI before candidates take part
- Purpose limitation: use candidate data only for the stated recruiting purposes, not for unrelated activities
- Data minimisation: collect only the information needed to assess suitability for the job
- Storage limitation: delete unsuccessful candidates' data within 6 to 12 months unless otherwise justified
- Accountability: document compliance through policies, assessments and training
Chapter 2: Legal basis for processing candidate data
The GDPR requires a legal basis for all processing of personal data. There are six possible bases: consent, contract, legal obligation, vital interests, public task and legitimate interests. This chapter shows which bases apply in recruiting and what each of them requires.
Legitimate interests as the primary basis
Most processing in recruiting relies on legitimate interests. Employers have an obvious legitimate interest in assessing candidates' suitability before making a hiring decision. This interest generally outweighs candidates' privacy interests, because candidates apply voluntarily and can reasonably expect to be assessed. Legitimate interests do, however, require a balancing test. It documents that the employer's interests outweigh candidates' data protection rights and that the processing is necessary for the legitimate interest, meaning that no less intrusive alternative exists.
For AI interviews, the balancing test must address whether AI assessment is necessary or whether traditional methods would be sufficient. Organisations can argue necessity on the grounds of efficiency (AI makes it possible to assess more candidates), consistency (AI applies uniform assessment standards) and quality (AI may be able to predict job performance more accurately). The balancing test should document these benefits and address candidates' privacy concerns through transparency, data minimisation and appropriate security. Anyone relying on legitimate interests must complete and document this analysis before processing begins – justification after the fact is not sufficient.
Contractual necessity as an alternative basis
Some processing can rely on contractual necessity – processing that is needed to assess whether an employment contract should be concluded. This basis covers information that is essential for assessing basic qualifications and suitability for the job. Contractual necessity is, however, narrower than legitimate interests. It does not support collecting information beyond what is strictly necessary for entering into the contract, and it does not justify processing after the application has been rejected (whereas legitimate interests can support short-term retention to defend against discrimination claims).
For AI interviews, contractual necessity is generally not sufficient as the sole basis, because AI assessment is useful but not strictly necessary for assessing the application – traditional methods are available as alternatives. Organisations should therefore rely primarily on legitimate interests. Contractual necessity can, however, support specific processing within the interview: collecting basic contact details, checking the right to work, assessing relevant skills and experience. The key difference: contractual necessity supports only processing that is genuinely required for entering into the contract, whereas legitimate interests can also justify broader processing that is useful but not strictly essential.
When consent is required
Consent is typically not the primary basis for processing in recruiting, because the employment context involves an inherent imbalance of power that undermines the voluntary nature of consent. Consent under the GDPR must be freely given, specific, informed and unambiguous – difficult to achieve when candidates fear that refusing will harm their application. Data protection authorities therefore consistently advise against relying on consent for core recruiting activities. Candidates should no more have to consent to basic assessment than to having their CV read.
Consent does become necessary, however, for processing that goes beyond what legitimate interests or contractual necessity support. Special categories of data (racial or ethnic origin, health information, religious beliefs, sexual orientation and so on) require explicit consent or one of the narrowly limited exceptions. Processing for purposes beyond recruiting – using candidate data for marketing, sharing it with third parties or adding it to candidate databases for future vacancies without clear initial notice – typically requires consent. Where consent is required, it must be given by a clear affirmative action (no pre-ticked boxes), separately from other agreements, and candidates must be genuinely free to refuse without this affecting their application for the specific position.
Restrictions on special categories of data
Special categories of data (racial or ethnic origin, health, genetic data, biometric data used for identification, religious beliefs, sexual orientation and so on) receive enhanced protection under the GDPR. Processing them generally requires explicit consent together with necessity for one of the purposes listed in Article 9(2). For recruiting, the relevant exceptions are those that permit processing necessary to meet obligations under employment law (for example assessing reasonable adjustments for disabilities) or that rest on a substantial public interest (for example equal opportunities monitoring with appropriate safeguards).
AI interviews carry particular risks here. Voice recordings are biometric data when they are used for identification. Accent analysis can reveal ethnic origin. Speech patterns can indicate disabilities. Video interviews can reveal racial characteristics, age or disabilities. Organisations must assess carefully whether their AI systems process special categories of data, even incidentally. If they do, they must establish a legal basis under Article 9(2), implement additional safeguards and make sure the processing is genuinely necessary. The safest approach is often to configure AI systems so that they avoid special categories of data altogether – audio-only interviews instead of video, content analysis without accent assessment, and disabling features that analyse demographic characteristics.
“In the employment context, it is doubtful whether consent is freely given. That is why legitimate interests, not consent, are the preferred legal basis in recruiting.”
Chapter 3: Transparency and candidate rights
The GDPR gives candidates extensive rights over their personal data and obliges organisations to provide comprehensive information about data processing. This chapter covers transparency obligations and candidate rights in AI recruiting.
Transparency requirements
Articles 13 and 14 require candidates to be given comprehensive information at or before the time their data is collected. The mandatory information includes: the identity and contact details of the controller, the contact details of the data protection officer, the purposes of the processing, the legal basis, the categories of data collected, the recipients of the data including third-party service providers, retention periods and the existence of candidates' rights. For AI interviews, organisations must additionally disclose that automated decision-making or profiling takes place, provide meaningful information about the logic involved and explain the significance and the envisaged consequences.
This information must be provided proactively and in clear, plain language that the average candidate can understand. It is not enough to hide mandatory information in long privacy policies that can only be reached through links that are hard to find. Good practice is a concise AI interview notice sent directly with the interview invitation that covers the key points: an AI will assess the answers, which factors the AI considers, whether decisions are fully automated or include human review, how candidates can request human intervention, and contact details for questions. The notice should link to a full privacy policy for candidates who want more detail, but the essential information must be presented concisely up front.
Right of access
Candidates have the right to obtain confirmation of whether their data is being processed and to access that data. This means providing copies of interview recordings, transcripts, AI assessments and scores, and any notes or assessments made by human recruiters. Organisations must respond to access requests within one month (extendable by a further two months for complex requests, with notice to the candidate). Requests must be fulfilled free of charge unless they are manifestly unfounded or excessive.
For AI interviews, access requests bring practical challenges. Candidates can ask for transcripts of long interviews, detailed explanations of the AI assessment methodology and information on how their answers compared with those of other candidates. Organisations should set up processes that allow them to fulfil these requests efficiently while protecting proprietary AI assessment logic and other candidates' confidential information. Technical measures such as candidate dashboards that proactively give access to interview recordings and scores can reduce formal access requests and show that the organisation takes transparency seriously.
Right to rectification and erasure
Candidates can request the correction of inaccurate data and, in certain circumstances, the erasure of their information. Rectification is straightforward – if candidate data contains errors, organisations must correct them promptly. Erasure (the right to be forgotten) is more complex. Candidates can request erasure when the data is no longer necessary for the purposes for which it was collected, when they withdraw consent (if consent was the legal basis), when they object and there are no overriding legitimate grounds, or when the processing was unlawful.
Erasure requests can, however, be refused where the processing is necessary to comply with legal obligations (for example retaining data needed to defend against discrimination claims) or to establish, exercise or defend legal claims. In recruiting, organisations typically keep unsuccessful candidates' data for 6 to 12 months in order to defend themselves against possible legal challenges to hiring decisions. Once this period has expired, continued retention requires a documented business justification. Organisations should set up automated deletion workflows that purge candidate data once retention periods expire, unless a specific reason supports longer retention.
Right to object and to restriction of processing
Candidates can object to processing based on legitimate interests. Organisations must then demonstrate compelling legitimate grounds that override the candidates' interests, or show that the processing is necessary for legal claims. In certain circumstances candidates can also request restriction of processing, in effect freezing their data until disputes are resolved. If candidates object to processing in the AI interview, organisations must assess whether they can evaluate the application by other means. If not, they may be able to continue processing but must document why their legitimate interests outweigh the candidates' objection.
Good practice is to accommodate objections wherever possible instead of forcing a confrontation. If a candidate objects to AI interviews, offering a traditional telephone screening as an alternative shows respect for the candidate's preferences, and qualifications can still be assessed. This approach reduces legal risk, improves the candidate experience and avoids the time and complexity of judging whether objections are valid. Organisations should set up clear escalation processes for applicants' requests concerning their data protection rights, with legal review of complex cases and documentation of the reasons for decisions.
- Transparency: give a clear AI interview notice with the invitation, covering the key processing details
- Access rights: respond to requests within one month and provide recordings, transcripts and assessments
- Rectification: correct inaccurate candidate data promptly on request
- Erasure: delete data when it is no longer necessary, typically 6 to 12 months after rejection
- Objection: offer alternative assessment methods when candidates object to AI processing
- Documentation: keep detailed records of how data protection rights requests were handled
Chapter 4: Requirements for automated decision-making
Article 22 GDPR provides special protection against automated decisions with legal or similarly significant effects. For AI interviews it must therefore be clear whether and how humans are involved in selection decisions. This chapter shows when Article 22 applies, which safeguards are required and what a compliant implementation looks like.
When Article 22 applies
Article 22 gives individuals the right not to be subject to a decision based solely on automated processing that produces legal effects concerning them or similarly significantly affects them. Recruiting decisions clearly have significant effects – whether someone gets a job has a major impact on their life. If AI interviews lead to automatic rejection without human involvement, Article 22 therefore applies. The decisive question is whether decisions are “solely” automated or include meaningful human review.
Meaningful human review requires that humans have the authority to depart from the automated assessment in their decisions, that they actively review AI recommendations instead of rubber-stamping them, that they consider factors beyond what the AI has assessed, and that they have the competence and information needed to evaluate AI assessments critically. It is not enough for a human simply to click “Approve” on AI-generated decisions. The human must genuinely evaluate the application, taking the AI assessment into account but not following it blindly. Where this meaningful human involvement exists, Article 22 probably does not apply; transparency and the other GDPR requirements nevertheless remain in place.
Required safeguards for automated decisions
Where decisions are solely automated (or human involvement is not sufficiently meaningful), Article 22 prohibits such processing unless there is explicit consent, the processing is necessary for the performance of a contract with appropriate safeguards, or it is authorised by law with suitable measures. In recruiting, explicit consent and statutory authorisation are typically not available, and contractual necessity is questionable for full automation. Organisations should therefore avoid purely automated rejection decisions or make sure that genuinely meaningful human review takes place.
Where automated decision-making takes place on a sound legal basis, organisations must implement safeguards: the right to human intervention in the decision, the right to express a point of view and contest the decision, regular accuracy and bias testing of the automated systems, and meaningful information about the decision logic. In essence, these requirements amount to human-in-the-loop for final decisions and explainable AI assessments. Organisations that use AI interviews should design their workflows so that the AI handles the first screening interview but humans make the final rejection decision; AI assessments serve as recommendations that humans can review and override.
Compliant AI interview workflows
The safest way to comply with Article 22 is to build explicit human decision points into AI interview workflows. For example: the AI conducts the initial interview and generates scores and recommendations. Candidates who score above a threshold (for example the top 40%) are passed automatically to a human recruiter for review. Candidates who score below the threshold are flagged for human review before any rejection. The recruiter reviews the AI assessment including the transcript and scores, considers additional factors such as application documents and role-specific requirements, and makes the final decision on whether to progress or reject the candidate, documenting the reasons.
This workflow ensures that no candidate is rejected automatically on the basis of the AI assessment alone. The human review step secures compliance with Article 22 while the efficiency benefits of AI are retained – recruiters spend their review time on the borderline cases flagged by the AI instead of interviewing every applicant themselves. Clear documentation of the human decision-making process demonstrates compliance and helps to defend against possible legal claims if decisions are challenged. Organisations should set up quality assurance processes to verify that human review actually takes place and has a meaningful influence on decisions, and is not merely a formal confirmation.
Explainability and contestability
Beyond human review, Article 22 requires meaningful information about the automated decision logic and the right to contest decisions. For AI systems that operate as black boxes, this is technically demanding. Organisations must reconcile the protection of their proprietary AI assessment methods with the legal transparency requirements. The solution typically lies in substantive explanations that do not disclose proprietary details: explaining in general terms which factors the AI considers (clarity of communication, technical knowledge, approach to problem-solving and so on) and which factors had the greatest influence on the individual candidate's assessment, without revealing detailed algorithms or scoring formulas.
Contestability requires clear processes by which candidates can challenge AI assessments they believe to be inaccurate. These can include the option of retaking the interview if candidates believe technical problems affected it, human review of contested assessments, or accepting alternative evidence of qualifications. Applicants must be able to see that they can question AI assessments, submit supplementary evidence and request a review. Organisations should document the contestation procedures in their privacy notices and respond to challenges promptly with a clear explanation of the outcome.
“Article 22 requires meaningful human review of AI rejections – automated screening recommendations must be reviewed by humans with the authority to override them.”
Chapter 5: Data minimisation and retention
The principles of data minimisation and storage limitation require collecting only necessary data and keeping it only for as long as is justified. This chapter applies both principles specifically to AI interviews: which data to collect, how long to keep it and when it must be deleted.
Determining what data collection is necessary
Data minimisation requires a critical analysis of which information is genuinely necessary for recruiting decisions. For AI interviews, necessary data typically includes: candidates' answers to job-relevant questions, the assessment of the competencies required for the role, technical performance indicators such as clarity of communication or problem-solving ability (where genuinely relevant to the position) and contact details for the next steps in the process. Data that is not necessary and should be minimised or excluded includes: demographic information that plays no part in the selection decision, personality traits unrelated to the job requirements, sentiment or emotional state (unless specifically job-relevant) and extensive personal background beyond employment history.
AI systems can extract vast amounts of information from interviews, but legal necessity covers only data that serves specific, documented recruiting purposes. Organisations should carry out a data mapping exercise that documents which information their AI interviews collect, why each category is necessary and how it is used in decisions. Any data that is collected but not actually used in the assessment should be dropped. If there are capabilities for analysing certain factors that are not job-relevant, those capabilities should be switched off. This disciplined approach secures compliance, reduces the risk of data breaches and respects candidates' privacy.
Retention periods for candidate data
Storage limitation requires deleting candidate data once the purposes have been fulfilled. Setting appropriate retention periods, however, means weighing competing considerations. Organisations need to keep data long enough to defend themselves against discrimination claims, which in most EU jurisdictions typically have to be brought within 3 to 6 months. Indefinite retention, on the other hand, cannot be justified once the time limits for possible legal claims have expired. Good-practice retention schedules typically provide for the following: Active candidates (still in the process): keep all data until the application process ends. Hired candidates: the retention rules for personnel files apply, typically 3 to 10 years depending on the jurisdiction. Rejected candidates: 6 to 12 months' retention to defend against possible legal claims, then deletion unless there is a specific, justified reason.
Retention periods should be communicated clearly to candidates in the privacy notices. Organisations cannot keep data indefinitely by pointing to vague business interests. Each category of data should have a documented justification for retention and a specified deletion deadline. On the technical side, this includes automated deletion workflows that purge data when retention periods expire; this reduces the risk of unauthorised extended retention. If candidates are placed in talent pools for future vacancies, this extended retention requires clear notice and a sound legal basis – typically explicit consent or a legitimate interest with documented justification.
Deletion procedures and verification
Effective deletion requires more than removing data from production databases. Genuine deletion includes: removing data from active systems, deleting backups containing candidate information once backup retention periods expire, purging data from disaster recovery and archive systems, instructing third-party processors to delete their copies, and documenting the completion of deletion for audit purposes. Organisations should set up systematic deletion processes and verify compliance through regular audits.
For AI interviews, deletion should cover interview recordings, transcripts, AI-generated scores and assessments, and all derived data produced by the interview analysis. It is not enough to delete the primary interview record while keeping scores or transcripts. Organisations should implement technical measures that flag data for deletion when retention periods expire and systematically purge all related records. Documenting the deletion processes and their verification provides important evidence of GDPR compliance and helps to defend against allegations of unauthorised data retention.
Special considerations for AI training data
Using candidate interviews to train or improve AI systems adds further complexity. If AI training was disclosed as a processing purpose when the data was collected, limited retention for this purpose may be justified. Organisations should, however, weigh carefully whether extended retention for AI training overrides candidates' privacy interests. Good practice is to obtain explicit consent for use in AI training, to anonymise training data so that it cannot be traced back to individual candidates, or to use synthetic data for AI training instead of real candidate information.
Once candidate data has been incorporated into AI training sets, later erasure requests are technically difficult – the data may have been absorbed into model weights in ways that prevent simple removal. This does not excuse non-compliance, but it shows how important careful planning is before candidate data is used for AI training. Organisations should assess whether the training purposes justify the intrusion into privacy and the complexity, or whether other training approaches (synthetic data, consenting research participants, public datasets) are more appropriate. If in doubt, decide in favour of data protection and limit retention.
- Data minimisation: collect only job-relevant information and switch off unnecessary AI analysis capabilities
- Rejected candidates: 6 to 12 months' retention to defend against possible legal claims, then mandatory deletion
- Hired candidates: the retention rules for personnel files apply, typically 3 to 10 years
- Deletion procedures: remove data from all systems, including backups and third-party processors
- AI training: obtain explicit consent or use anonymised/synthetic data instead of candidate information
- Documentation: record the justification for retention and the completion of deletion
Chapter 6: International data transfers
If AI interview systems transfer data to countries outside the European Economic Area, additional legal requirements apply. This chapter covers the restrictions on international transfers, the transfer mechanisms available and the particular issues raised by common AI interview architectures using US-based cloud providers.
Why international transfers matter
The GDPR restricts the transfer of personal data to countries outside the EEA unless adequate protection is ensured. The restriction applies when candidates' interview data is stored on servers outside the EEA, processed by service providers outside the EEA or accessed by staff at locations outside the EEA. Many AI interview platforms use cloud infrastructure from providers such as AWS, Google Cloud or Microsoft Azure, which may involve international transfers. Even if the data is held in EU data centres, remote access to the systems by technical support staff in the US is an international transfer that requires a legal mechanism.
The complexity arises because the GDPR does not prohibit international transfers but requires specific legal mechanisms that ensure data protection equivalent to EU standards. The Schrems II ruling invalidated the Privacy Shield in 2020. Since 10 July 2023 a new adequacy decision has applied in the form of the EU-US Data Privacy Framework: transfers to US providers certified under the framework are permitted without additional mechanisms. For US providers that are not certified, and for other third countries without an adequacy decision, organisations must continue to use standard contractual clauses or binding corporate rules with additional safeguards. The EU General Court upheld the framework in September 2025; an appeal before the Court of Justice is pending.
Standard contractual clauses (SCCs)
The European Commission has approved standard contractual clauses – standardised agreements between data exporters (EU organisations) and data importers (service providers outside the EU) that set out data protection obligations. SCCs are the usual mechanism for transfers to providers not covered by an adequacy decision, such as US providers without certification under the Data Privacy Framework. To rely on SCCs, organisations must incorporate the approved SCC wording into their contracts with service providers, carry out transfer impact assessments that examine whether the laws of the destination country undermine the protection of the SCCs, implement supplementary measures (encryption, pseudonymisation, access controls) to address the risks identified, and document the assessment and the decisions.
For AI interviews running on US cloud infrastructure, the transfer impact assessment should address whether US surveillance laws (in particular FISA Section 702 and Executive Order 12333) could give government bodies access to candidate data. If the service provider is subject to these laws and the candidate data is not protected by encryption or other technical measures, additional safeguards are required. Practical approaches: requiring providers to store EU candidate data exclusively in EU data centres, introducing encryption for which the provider holds no decryption keys, restricting the provider's access to the data in order to minimise surveillance risk, and obtaining detailed assurances about the provider's data processing practices and its handling of government data requests.
Adequacy decisions and alternative mechanisms
For some jurisdictions the European Commission has issued adequacy decisions that allow data to flow freely without additional mechanisms. These currently include Andorra, Argentina, Canada (commercial organisations), the Faroe Islands, Guernsey, Israel, the Isle of Man, Japan, Jersey, New Zealand, South Korea, Switzerland, the United Kingdom, Uruguay and the United States for companies certified under the EU-US Data Privacy Framework. Organisations that use service providers in these locations can transfer data without SCCs; the basic data protection principles continue to apply.
Other transfer mechanisms are binding corporate rules for intra-group transfers within multinational companies, derogations for specific situations (explicit consent, contractual necessity, important public interest) and approved codes of conduct or certification mechanisms. These alternatives, however, either do not fit most AI interview scenarios (BCRs require substantial investment that can only be justified for large multinationals) or are not sufficient for regular business transfers (derogations are intended for exceptional situations, not systematic transfers). For most organisations, SCCs remain the practical transfer mechanism.
Practical recommendations for a compliant architecture
Organisations can keep the complexity of international transfers low by making careful architecture decisions for their AI interviews. Ideally, choose providers that guarantee EU data residency – the data is held exclusively in EU data centres and accessed only by staff based in the EU. This avoids international transfers or reduces them to a minimum. If US-based providers are necessary, implement technical measures that protect the data even if unauthorised parties gain access to it: end-to-end encryption for which only your organisation holds the decryption keys, pseudonymisation or anonymisation of the data where possible, and access controls that restrict who can view identifiable candidate information.
Contractual measures are just as important. Make sure that your service agreements contain the following: approved standard contractual clauses, clear commitments on where the data is located, restrictions on onward transfers without consent, transparency obligations for government access requests and audit rights to verify data processing practices. Organisations should document their transfer assessments, the safeguards used and their ongoing monitoring. This documentation demonstrates that data protection has been implemented and serves as a defence if transfers are challenged. As the law on international data transfers continues to develop, organisations should review their transfer arrangements regularly and adjust them where necessary in order to stay compliant.
- International transfers: a legal mechanism is required when data leaves the EEA, including cloud storage and remote access
- Standard contractual clauses: the most common mechanism; requires a transfer impact assessment and supplementary measures
- Transfer assessments: assess destination-country risks, in particular US surveillance laws
- Technical safeguards: encryption, pseudonymisation and EU data residency minimise transfer risks
- Provider selection: prefer providers that offer EU data residency and EU-based support staff
- Documentation: keep records of transfer assessments, SCCs and supplementary measures
Chapter 7: Implementation checklist and best practices
This chapter arranges the measures from the previous chapters into checklists for implementation and operation. Use them to review your current practices, identify gaps and develop plans for remedying shortcomings.
Pre-implementation checklist
Before you deploy AI interviews, complete these basic steps: Carry out a data protection impact assessment (DPIA) that documents the processing activities, necessity, risks and mitigation measures. Establish the legal basis for processing, typically legitimate interests with a documented balancing test. Draft a comprehensive privacy notice that covers all the requirements of Article 13 and the specific disclosures for AI interviews. Design an AI interview workflow that ensures meaningful human review of rejection decisions (compliance with Article 22). Negotiate data processing agreements with AI interview providers, including standard contractual clauses where necessary.
In addition, implement technical safeguards: Configure AI systems to minimise data collection and switch off unnecessary analysis capabilities. Set up encryption for data in transit and at rest. Create access controls that restrict who can view candidate information. Set up automated deletion workflows based on documented retention periods. Develop security incident response procedures for potential data breaches. Train recruiting staff in the GDPR requirements and in the proper use of the systems. These steps secure compliance from the start and avoid the need for corrections after the fact.
Data protection in ongoing operation
GDPR compliance is not a one-off implementation but an ongoing operational task. Establish these regular practices: Quarterly bias audits that test AI interview results for adverse impact on protected groups. Annual DPIA updates that reflect changes in processing activities or the risk environment. Regular security audits and penetration tests of systems holding candidate data. Systematic monitoring of retention periods with automated deletion when they expire. Prompt responses to applicants' requests concerning their data protection rights within the one-month time limit. Upkeep of documentation including records of processing, impact assessments and the handling of rights requests. Legal monitoring to track GDPR guidance, enforcement action and regulatory developments.
In addition, implement continuous improvement processes. Review applicants' feedback on AI interviews and on privacy transparency. Analyse metrics on applicants' requests concerning their data protection rights in order to identify potential compliance problems. Carry out periodic gap assessments that compare current practices with evolving best practice. Update privacy notices and processes as regulations change. This proactive approach to data protection reduces enforcement risk and demonstrates good-faith efforts to protect candidates' privacy – a factor that regulators take into account when setting penalties for breaches.
Handling data protection rights requests
Set up clear procedures for handling applicants' requests concerning their data protection rights efficiently and compliantly: Designate responsible staff with appropriate training and decision-making authority. Create contact channels for requests (email address, web form, prominently published contact details). Implement identity verification procedures that prevent unauthorised access to candidate data. Define response deadlines supported by templates: acknowledge requests within 3 to 5 days, provide substantive responses within one month, and explain any extension within the initial one-month period. Build the technical capability to extract candidate data efficiently for access requests.
For complex cases, set up escalation procedures: Legal review in ambiguous situations or where refusing a request is being considered. Sign-off by senior management for unusual cases or those that could set a precedent. Documentation requirements that capture the details of the request, the analysis, the decisions and the reasons for the response. These procedures ensure consistent, compliant handling of data protection rights requests and bring to light potential systemic problems revealed by patterns in the requests.
Data protection beyond the minimum requirements
The previous sections deal with the compliance requirements. Leading organisations go beyond the minimum legal obligations in order to build candidates' trust. Best practices include: Proactive transparency through candidate dashboards that show interview recordings, transcripts and AI assessments without the need for formal access requests. A privacy-by-design approach that takes data protection into account from the initial system design and not as an afterthought. Candidate choice, with alternatives to the AI interview for those who prefer traditional methods. Investment in explainability that gives meaningful insight into AI assessments beyond the minimum required disclosures.
Then there are privacy-enhancing technologies such as differential privacy in AI training and federated learning architectures that minimise central data collection. Regular privacy audits by external experts as independent verification. Transparent reporting on the use of recruiting data, on data protection practices and on the handling of applicants' requests concerning their data protection rights. These measures make the handling of data understandable for applicants. They can strengthen trust in the company and matter to applicants who look at data protection when choosing an employer.
- Before implementation: complete the DPIA, establish the legal basis, draft privacy notices, implement technical safeguards
- Ongoing operation: quarterly bias audits, annual DPIA updates, systematic monitoring of deletion
- Data protection rights requests: one-month response deadline, identity verification, documented reasons for decisions
- Security: encryption, access controls, regular audits, incident response procedures
- Best practices: proactive transparency, candidate choice, privacy by design, external audits
- Documentation: keep comprehensive records that demonstrate the implementation of data protection and good faith
White paper and further material
Download the full white paper with all diagrams, templates and resources.
See Talky with your jobs?
30 minutes, with an example from your recruiting.