The European Data Protection Board (EDPB) has chosen automated decision-making, including AI-assisted recruitment systems, as the focus of its sixth coordinated enforcement action. Supervisory authorities will join on a voluntary basis soon, and the action itself launches in 2027. The focus is clear: the right not to be subject to a decision based solely on automated processing, and the related right of access, in the context of automated candidate screening tools.
This narrow focus highlights where compliance questions will arise. Here are the mistakes most likely to surface.
Why These Mistakes Keep Happening
Recruitment tools often enter a business through HR or talent acquisition, not legal. processor demos emphasize speed and candidate ranking, while Article 22 of the GDPR is rarely mentioned. By the time compliance teams see the contract, the tool is already part of the hiring process.
There's also a definitional trap. Teams hear "automated decision-making" and imagine a system that sends rejection emails without human input. Most screening tools don't operate that way. They score, rank, or filter, and a recruiter reviews the output. Whether that review is meaningful is the question regulators will ask.
Mistake 1: Treating a Recruiter's Sign-Off as Human Involvement
Why it happens: The tool produces a shortlist, a human clicks "proceed," and the team concludes the decision isn't solely automated.
Consequence: If the recruiter lacks the authority, information, or time to deviate from the tool's output, the review may be a formality. Article 22(1) applies to decisions based solely on automated processing, and a rubber-stamp step won't necessarily take you outside it. A supervisory authority examining your process can ask how often reviewers overrule the system.
Fix: Document what the reviewer sees, what they're empowered to change, and whether they have access to the underlying candidate information beyond the score. Track override rates. If reviewers never disagree with the tool, ask why before a supervisory authority does.
Mistake 2: Skipping the Lawful Basis Analysis for Article 22
Why it happens: Teams settle the Article 6 lawful basis for recruitment (often legitimate interests or steps prior to a contract) and assume that covers the automation.
Consequence: Where Article 22 applies, the decision must fall within one of the exceptions in Article 22(2): necessary for entering into a contract, authorized by Union or Member State law, or based on explicit consent. "Necessary" is interpreted narrowly. Efficiency alone is a weak argument when a human could do the screening. Relying on explicit consent in an employment context raises its own problem, because candidates rarely have a genuine choice.
Fix: Run a separate Article 22 assessment alongside your Article 6 analysis. Record which exception you rely on and why the automation is necessary, not merely convenient. If you can't support it, redesign the process so a person makes the decision.
Mistake 3: Writing Transparency Notices That Describe the Tool, Not the Logic
Why it happens: Vendors treat their scoring models as proprietary, so privacy notices stay generic: "we may use automated tools to assist recruitment."
Consequence: Articles 13(2)(f) and 14(2)(g) require meaningful information about the logic involved, the significance, and the envisaged consequences of automated decision-making. A generic sentence doesn't meet your transparency obligations. It also leaves you unable to answer a candidate who asks how they were assessed.
Fix: Require vendors contractually to give you enough detail to explain the main factors in plain language. Put that explanation in your candidate-facing notice at the application stage, not buried in a general policy.
Mistake 4: Handling Candidate DSARs as an Afterthought
Why it happens: Recruitment data sits in the applicant tracking system, the processor's platform, and email. Nobody owns the end-to-end picture, and rejected candidates aren't a routine DSAR population.
Consequence: The EDPB's action covers the right of access in this context, so expect authorities to test whether you can produce scores, rankings, and the information used to generate them. A data subject access request (DSAR) that returns only a CV and an application form will look incomplete when the decision was driven by a model output.
Fix: Map every place candidate data and derived outputs are stored, including inside the processor's systems. Add DSAR cooperation duties and response timelines to your processor agreement under Article 28. Test the process with a mock candidate request before launch.
Mistake 5: Assuming the Processor Relationship Is Someone Else's Problem
Why it happens: The processor is the specialist, so teams assume the processor handles compliance. Where the processor is based outside the EU, transfer questions get pushed to a later phase.
Consequence: You remain the controller and carry the Article 22 and transparency duties. On transfers, the mechanism matters. An adequacy decision, such as the one covering South Korea, gives you a clean route to a recipient there, and the Commission's first review concluded that the country continues to provide an adequate level of protection. But adequacy covers a specific country and framework, not every processor everywhere. For intra-group processor transfers, Binding Corporate Rules for Processors are an option, and the EDPB's final recommendations now apply to all BCR-P holders, who should align during their application or at their 2028 annual update, depending on their situation.
Fix: Identify the transfer mechanism for each processor in the recruitment chain and record it. If a processor relies on BCR-P, check that its rules reflect the final recommendations. Confirm sub-processor chains too.
Prevention Checklist
- Inventory every screening, ranking, or filtering tool used in hiring, including features switched on by default.
- Test each tool against Article 22(1): is human review meaningful, with authority and information to override?
- Document the Article 22(2) exception relied on, separately from the Article 6 lawful basis.
- Complete a data protection impact assessment where profiling of candidates is systematic or involves new technology.
- Update candidate notices with meaningful information about the logic, significance, and consequences.
- Build and test a DSAR route that captures scores and derived outputs held by processors.
- Review processor agreements for Article 28 terms, DSAR support, and sub-processor controls.
- Record the transfer mechanism for each processor: adequacy, BCR-P, or other safeguards.
- Track reviewer override rates and sample decisions periodically.
- Assign one named owner across HR, legal, and procurement for the recruitment tooling.
Authorities will join the coordinated action over the coming weeks and launch it in 2027. That gives you a window to fix the process before questions arrive.





