Skip to content
senn-techsenn-tech
Back to blog
Strategie
AI explained: models, market, news2026-11-22· By Franz Senn

AI agents in HR: when the works council must consent (§ 96, § 96a ArbVG, § 87 BetrVG)

Anyone in Austria or Germany who introduces an AI agent for personnel tasks meets obligations in two places: at the works council, and at the question whether the agent decides about people. This article sets the statutes side by side, as we opened them on 22 November 2026. It is not legal advice. For a specific company the case belongs with a lawyer specialising in employment law.

An HR agent here means an AI system that handles personnel-related tasks: filing applications, proposing appointments, guiding new hires through their first week, answering internal questions from policies. It needs works council consent when it processes employee data or makes behaviour and performance measurable. It becomes high-risk under the AI Act when it evaluates, ranks or screens applicants or employees. The line depends on what the system is used for, not on the model behind it.

Where the hard line on decisions about people lies

Art. 22(1) GDPR reads, in the German text on EUR-Lex (our translation): the data subject has the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning them or similarly significantly affects them. Paragraph 2 lists the exceptions: performance of a contract, a legal provision with safeguards, or explicit consent.

The WKO guidance on the GDPR for SMEs names the evaluation of applicants by an AI tool as an example and states that a pro forma sign-off by a human at the end is not a "human in the loop". In case C-634/21 of 7 December 2023 the Court of Justice held that even an automatically generated probability value can be a decision within the meaning of Art. 22 if a third party relies on it decisively. That case concerned a credit agency. We use it here only as a warning: a preparation that the decision-maker relies on is easily more than a preparation.

The AI Act draws the same line. Annex III point 4 of Regulation (EU) 2024/1689 lists as high-risk AI systems that screen or filter applications and evaluate candidates. Under Art. 6(3), such a system is not high-risk if it performs only a narrow procedural task or a preparatory task. According to the last subparagraph, that never applies when the system performs profiling.

The Commission's reading is in its draft guidelines on classification of 19 May 2026, which are not binding. We read only the draft and did not check whether a final version exists. As exceptions under Art. 6(3) it lists: recognising and organising CV data in a searchable database, scheduling interviews with reminders, and personalised acknowledgement emails. High-risk, by contrast, are suitability scores, rankings ("top 5"), fit categories such as "high fit", and competence profiles that condition shortlisting, even where recruiters can overrule them. An onboarding assistant that provides policies and training dates falls, according to the draft, outside point 4(a), but may fall under point 4(b) if it feeds performance evaluations or monitors workers.

The Labour Constitution Act (ArbVG) has two consent provisions. § 96 para. 1 covers, in our translation, "the introduction of control measures and technical systems for controlling employees, where these measures (systems) affect human dignity" (no. 3) and personnel questionnaires that go beyond general particulars and professional qualifications (no. 2). Such measures need the works council's consent "to be legally effective". § 96 contains no rule allowing the consent to be replaced.

§ 96a para. 1 is headed "Ersetzbare Zustimmung" (replaceable consent). No. 1 covers "systems for the automated collection, processing and transmission of personal data of the employee that go beyond general particulars and professional qualifications". No consent is needed where the use does not exceed obligations arising from law, collective agreements or the employment contract. No. 2 covers systems for appraising employees which collect data not justified by the operational use. Under para. 2 the arbitration board (Schlichtungsstelle) can replace the consent.

Alongside stands § 91 para. 2: the employer must tell the works council which types of personal employee data it records by automated means and which processing and transmission it plans. The Vienna Chamber of Labour classifies this the same way in an interview of 20 October 2026 and points to § 96a for AI systems used for appraisal. The wording says "Arbeitnehmer" (employee). We did not check whether applicants are covered.

What applies in Germany

§ 87(1) no. 6 BetrVG gives the works council a co-determination right on the "introduction and use of technical devices intended to monitor the behaviour or performance of employees" (our translation). There is no threshold such as human dignity, as in the Austrian text. If no agreement is reached, the conciliation committee (Einigungsstelle) decides under para. 2.

The statute names three further points for AI. § 90(1) no. 3 obliges the employer to inform the works council in good time about the planning of work procedures "including the use of artificial intelligence". § 80(3) sentence 2 deems an expert necessary where the works council has to assess the introduction or use of AI. § 95(2a) applies the rules on selection guidelines also where AI is used in drawing them up.

Austria and Germany side by side

RuleAustria (ArbVG)Germany (BetrVG)Source
Works council from5 permanent employees entitled to vote5 permanent employees entitled to vote, 3 of them eligible§ 40(1) ArbVG, § 1(1) BetrVG
Control technologyConsent where human dignity is affectedCo-determination for devices intended to monitor behaviour or performance§ 96(1) no. 3 ArbVG, § 87(1) no. 6 BetrVG
Personnel data systemsConsent under § 96a(1) nos. 1 and 2Consent for personnel questionnaires and general appraisal principles (§ 94); a general consent right for personnel data systems is not in the paragraphs we read§ 96a ArbVG, § 94 BetrVG
If no agreementArbitration board replaces consent under § 96a, not provided in § 96Conciliation committee replaces the agreement under § 87§ 96a(2) ArbVG, § 87(2) BetrVG
InformationNotice of data types and processingInformation on planning, expressly including AI§ 91(2) ArbVG, § 90(1) no. 3 BetrVG
AI expertnot checked by usdeemed necessary§ 80(3) BetrVG
Selection guidelinesnot checked by usConsent, also where AI is used§ 95(1), (2a) BetrVG

Unit: legal provision. Date: 22 November 2026, version as retrieved. Source: statute text on ris.bka.gv.at and gesetze-im-internet.de, no case law and no commentary. Art. 22 GDPR and the AI Act apply equally in both countries.

The three tasks one by one

For the application inbox: filing, completeness checks, proposing dates and sending confirmations match the examples the Commission's draft lists as procedural tasks. If the system evaluates the documents, it is high-risk and must also be checked against Art. 22 GDPR. In our reading the works council comes in here mainly through the personnel questionnaire (§ 96(1) no. 2 ArbVG, § 94 BetrVG) and through data of existing employees. That is a deduction from the wording, not a finding.

For onboarding: the draft says a schedule and reminders are not a recruitment case. Data of new hires is still employee data, and reporting their progress to a manager would be an appraisal.

For the knowledge assistant: it answers questions from policies. The log is the critical part. Who asks what, how often and when is behaviour, which is exactly what § 96(1) no. 3 ArbVG and § 87(1) no. 6 BetrVG address. Personnel files and salary lists do not belong in the index. We run separate Qdrant collections per source (project documentation, as of 22 November 2026). Whether access is separated per user we have not checked, so we do not claim it.

Deadlines and what already applies

Article 4 of the AI Act (AI literacy) has applied since 2 February 2025. Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026, reworded it: providers and deployers take measures to support the development of AI literacy, without guaranteeing a particular level for anyone. The background is in our article on Article 4 and proof of compliance in Austria. The high-risk obligations for Annex III systems apply from 2 December 2027 under that regulation, for Annex I systems from 2 August 2028. With them comes Art. 26(7) AI Act: employers must inform employee representatives and the affected employees before putting a high-risk system into use at the workplace. Inferring emotions in the workplace has been prohibited since 2 February 2025, except for medical or safety reasons (Art. 5(1)). Co-determination and the GDPR apply without a transition period.

Symbol of the European Union, a ring of twelve stars
The Commission published its high-risk classification guidance as a consultation draft in May 2026. (Quelle: EU Digital Strategy)

What we recommend

Start with a knowledge assistant over policies, with no personal data and no usage log with names. Set it down in writing that the agent neither evaluates, ranks nor screens out applicants or employees, and put that rule into the description of the system. Where a works council exists, inform it before the pilot and clarify whether § 96 or § 96a ArbVG, or § 87 BetrVG, applies. Anyone who wants AI to assess suitability later should plan for the high-risk obligations from December 2027 and take specialist advice. For building a company-owned assistant see our services page; more articles are collected under AI in production.

Further reading

Questions?
Does an Austrian company need works council consent for a knowledge assistant?+

It depends on the data and on how the system is evaluated. § 96a para. 1 no. 1 ArbVG requires consent for systems that process employees' personal data beyond general particulars. § 96 para. 1 no. 3 applies to control systems that affect human dignity. An assistant that only searches policies and keeps no usage logs with names touches both less. Once questions are stored per person, involving the works council is the safe route. This is not legal advice.

May an AI agent pre-sort incoming applications?+

Not if it scores, ranks or screens them out. The Commission's draft guidelines on high-risk classification treat suitability scores, rankings and competence profiles as high-risk under Annex III point 4(a). Recognising and organising CV data, scheduling interviews and sending acknowledgements are listed as exceptions under Art. 6(3). Art. 22 GDPR adds a further limit when no human genuinely takes part in the decision.

When do the high-risk obligations for AI in HR apply?+

For Annex III systems, which include employment and workers management, the obligations apply from 2 December 2027 under Regulation (EU) 2026/1744. That regulation was published in the Official Journal on 24 July 2026 and entered into force three days later. The AI literacy duty in Article 4 already applies, as do Art. 22 GDPR and the co-determination rights.