EU AI Act Guide
Is AI Resume Screening High-Risk Under the EU AI Act? (Yes — Here's What That Means)
Last updated: 2026-07-09
Short answer: yes. AI systems used for recruitment or selection - placing targeted job ads, analysing and filtering applications, evaluating candidates - are listed in Annex III, point 4(a) of the EU AI Act. That makes them high-risk systems, with current Commission implementation material pointing to 2 December 2027 as the latest application date for Annex III high-risk rules.
This affects two groups differently: the HR-tech vendors who build screening tools (providers), and the employers who use them (deployers). Both have duties; they are not the same duties.
Not sure which tier applies to you? Check your system in 2 minutes — free and rule-based.
Quick shortcut
Already know you need documents? Skip straight to the Doc Pack — your classification is done for you.
Get Doc Pack ($229–$499)Why resume screening is Annex III
Employment decisions shape people's livelihoods, which is exactly the kind of fundamental-rights impact the Act targets. Annex III point 4 covers AI intended for recruitment or selection of natural persons — advertising, screening, filtering, evaluating — plus systems for promotion, termination, task allocation and monitoring at work.
It does not matter whether a human makes the final call. If the AI ranks, filters or scores candidates along the way, the system is in scope. The 'human in the loop' matters for the oversight obligations — it does not remove the classification.
Provider duties (you build the screening tool)
Providers carry the heavier load: a risk management system (Art. 9), data governance for training and testing data including bias examination (Art. 10), Annex IV technical documentation (Art. 11), event logging (Art. 12), instructions for use (Art. 13), human oversight design (Art. 14), accuracy and robustness declarations (Art. 15), then conformity assessment, CE marking and EU database registration (Art. 43, 47–49).
The bias-examination duty in Article 10 deserves special attention for hiring tools — your training data must be examined for possible biases likely to affect health, safety or fundamental rights, and you must document what you found and did.
Deployer duties (you use someone else's screening tool)
Employers using AI screening must use the system per its instructions, assign trained human oversight, monitor operation, keep logs (Art. 26), and inform workers and candidates that AI is in use. Public-sector employers and some others must also complete a Fundamental Rights Impact Assessment before first use (Art. 27).
Practical takeaway: when your screening vendor cannot produce their Annex IV documentation, that is now your problem too — procurement teams have started asking, and deployers cannot meet their own duties with an undocumented system.
The Art. 6(3) exemption — narrow and documented
A system that only performs a narrow procedural task, or purely preparatory work, within an Annex III area may escape the high-risk classification under Article 6(3). A tool that merely deduplicates applications or converts CVs to a standard format is plausibly exempt; anything that ranks or scores candidates is not. Claiming the exemption requires a documented assessment registered before market placement — it is not something you can assert after the fact.
You now know what's required. Rather not write it from scratch?
The AI Act Documentation Pack delivers your risk classification memo, Annex IV technical documentation, model card, AI usage policy and compliance roadmap — generated from a 15-minute questionnaire after checkout confirmation, from $229. 14-day refund if the delivered pack materially differs from the contents described at purchase. Not legal advice.
See the Doc Pack →Frequently asked questions
We only use AI to parse CVs into fields, not to rank candidates. High-risk?
Pure format conversion with no evaluative function is the strongest case for the Art. 6(3) narrow-task exemption — but the exemption must be documented and defensible. If any scoring, matching or filtering happens, assume high-risk.
Does a human making the final hiring decision change the classification?
No. Human review is an obligation for high-risk systems (Art. 14), not an exit from the category. The system is classified by its intended purpose, not by who clicks the last button.
We're a US HR-tech company with EU customers. Are we in scope?
Yes. The Act applies when a system is placed on the EU market or its output is used in the EU — your customers' location pulls you in, regardless of where you are incorporated.
Related guides
EU AI Act Deadlines: The Complete 2025-2028 Timeline
Every EU AI Act deadline from February 2025 to August 2028 - what applies when, to whom, and what to do before 2 December 2027.
AI Act Compliance Tools Compared (2026): Platforms, Packs & DIY
Honest comparison of EU AI Act compliance options — GRC platforms like Vanta and Drata, documentation packs, and doing it yourself. Sized by risk, systems and workflow needs.
Annex IV Technical Documentation: Structure, Template & Examples
What EU AI Act Annex IV technical documentation must contain, section by section — and how to produce a structured first draft without a custom law-firm drafting engagement.
EU AI Act Compliance Checklist for SaaS Founders (2026)
A practical six-step EU AI Act compliance checklist for SaaS and AI teams - classify your systems, track the 2 August 2026 transparency date and 2 December 2027 Annex III date, and produce the documents buyers ask for.
About these guides
Written by the TrustPacket team. We track Regulation (EU) 2024/1689, Commission guidance, and enforcement developments to keep these guides current. Not legal advice — when in doubt, consult qualified counsel.
Last updated: 2026-07-09 · Content review: 11 July 2026 · Questions? Email us
This is informational editorial content based on Regulation (EU) 2024/1689 and cited official implementation material. It is not legal advice, not a conformity assessment, and does not create a lawyer-client relationship. Application depends on facts this guide cannot verify. Consult qualified counsel for a binding assessment.