AI Hiring Stage by Stage: What Is High-Risk (EU, Germany, US)

AI hiring means using software to support or decide steps of the hiring process: writing and placing job ads, finding and matching candidates, screening CVs, scheduling, and interviewing by chat or voice. Whether a step is allowed, regulated or high-risk depends on the stage and on one question: does the tool only prepare, or does it select? In the EU, AI that places targeted job ads, filters applications or evaluates candidates is high-risk under Regulation (EU) 2024/1689, Annex III no. 4 (a), with obligations from 2 December 2027. In New York City, automated employment decision tools already need an annual bias audit. This guide walks through the process stage by stage. It is general information, not legal advice.

The hiring process, stage by stage

Most guides on AI hiring list risks in general. What a team needs is a map by stage: what the tool does there, how it is classed, and what you must have ready.

Sources: Annex III no. 4 (a) covers AI "intended to be used for the recruitment or selection of natural persons", in particular to place targeted job advertisements, to analyse and filter applications and to evaluate candidates (Annex III). The New York City rules are described by the city's consumer agency: no use of an automated employment decision tool unless it was subject to a bias audit within one year, public information on the audit, and notice to candidates; enforcement began on 5 July 2023 (NYC DCWP). The rule applies to people who have applied for a specific position, not to scanning resume banks or outreach (DCWP FAQ). Check state rules yourself; several US states regulate AI in hiring and the details change.

Is a step "assistance" or "decision"?

This is the question that decides most cases, and the answer is less comfortable than vendors suggest. A tool that proposes and a recruiter who signs off does not automatically mean a human decision. The Court of Justice held in the SCHUFA case (C-634/21, 7 December 2023) that a score can itself be a decision where a third party relies heavily on it. Use three tests:

1. Could the recruiter realistically disagree? If they see only a rank and a number, probably not.
2. Does the tool filter before a human sees anything? Candidates who are never shown to a person were decided by the system.
3. Can you reconstruct why? If the reason for a rating is unavailable, you cannot defend it (see burden of proof below).

Our guide Can AI reject candidates automatically? goes deeper on Art. 22 GDPR.

When do the EU duties start, and what do they require?

The Digital Omnibus, Regulation (EU) 2026/1744, moved the application of the high-risk requirements for Annex III systems to 2 December 2027 (originally 2 August 2026). Banned practices, including emotion recognition at work, have applied since 2 February 2025; the transparency duty (Art. 50) has applied since 2 August 2026. As deployer you must use the system as instructed, assign human oversight to competent people, keep the logs, and inform workers and their representatives before a high-risk system goes into use at work (Art. 26). Choose a vendor who can give you instructions, documentation and logs, and ask for them before you sign. The full legal walk-through for DACH is in our compliance overview.

Germany: works council and equal treatment

• § 87 (1) no. 6 BetrVG: co-determination on technical systems designed to monitor behaviour or performance. Recruiting platforms usually log reviewer actions, which is enough to trigger it. The conciliation board decides if the parties disagree (§ 87 (2)).
• § 95 (2a): selection guidelines require consent, also where AI is used to create them. Selection guidelines for hiring are exactly what a screening tool encodes.
• § 22 AGG: if the applicant shows indications of disadvantage, the employer must prove there was none. Compensation for a rejection can reach up to three monthly salaries where the person would not have been hired anyway (§ 15 (2) AGG). An opaque model makes the proof hard.

What does AI screening cost? A worked example

At Sprad, screening one application with CV check (2 credits), candidate updates (9) and a voice interview (30) is 41 credits. At the list price of 0.10 € per credit that is 4.10 € per application; at the Atlas Base price of 0.09 € it is 3.69 € (source: pricing, status 29 October 2026). For 300 applications on a position: about 1,107 to 1,230 €. If you start with the CV check alone (2 credits, 0.18 to 0.20 €), 300 applications cost 54 to 60 €. Compare this with the recruiter hours for the same volume at your own hourly cost. These are our prices as a vendor; other tools price differently.

Before you start: a short checklist

1. Write down for each stage which tool does what, and whether it filters or only supports.
2. Define must-have criteria that are job-related. Experience shows that results get worse when must-haves are set as nice-to-haves.
3. Request the vendor's documentation, logs and instructions for use.
4. Run a data protection impact assessment with your data protection officer.
5. Involve the works council early; bring the function list and the draft works agreement.
6. Test on a real open position with human review and compare to the manual shortlist.
7. Check outcomes by group at fixed intervals and keep the records.

In our conversations with recruiting teams, data protection is usually a checkpoint, not a blocker, once hosting, retention, deletion and consent are answered in concrete terms.

FAQ

Is AI hiring legal?

Yes, within limits. In the EU and Germany it is allowed subject to data protection, equal treatment and, for the high-risk class, the AI Act duties from 2 December 2027. Some practices, such as emotion recognition of employees, are prohibited. In the US, federal discrimination law applies and several states and cities add rules.

Can AI reject candidates?

Not on its own where Art. 22 GDPR applies. Keep a person in the loop who can really decide.

Is AI hiring biased?

It can be. Models learn from past data and from proxies. Define job-related criteria, test for different outcomes by group and keep logs.

Which AI hiring tools exist?

See the comparison of AI recruiting software. Sprad itself offers an AI recruiting platform with AI sourcing and CV screening and voice interviews. Sprad is a vendor, so treat this as one view among several.

How is this different from AI in HR in general?

Hiring is one area. The wider picture, including performance and people analytics, is in AI in HR.

As of 8 November 2026. General information, not legal advice. Legal references checked against the official texts and the city's published guidance. Sprad is a vendor of recruiting software; have the specific use reviewed by your data protection officer and employment counsel.