Do you have to inform employees and their representatives before deploying AI at work? Yes, and in most EU countries the duty already exists today under national labour law and the GDPR, while the EU AI Act adds its own obligation in Article 26(7), which, based on our reading, applies to high-risk systems only and from 2 December 2027. The practical question is therefore not whether to inform, but which rule triggers first and what the notice must contain.
Stand: 2026-10-16
This page covers the EU layer and the two national systems we work in most, Germany and Spain, because that is where the detail lives. The AI Act was written to sit on top of national worker information law, not to replace it.
The EU AI Act: Article 26(7) and its limits
Article 26(7) of Regulation (EU) 2024/1689 requires deployers who are employers to inform workers' representatives and the affected workers, before putting a high-risk AI system into service or using it at the workplace, that they will be subject to its use. Where applicable, the information follows the rules and procedures of Union and national law on informing workers and their representatives, so the national channel described below is also the AI Act channel.
The scope is narrower than the headline suggests. The duty attaches to high-risk systems, and the workplace cases are listed in Annex III, point 4: systems used for recruitment or selection, in particular targeted job advertisements, filtering applications and evaluating candidates (point 4(a)), and systems used to make decisions on terms of employment, promotion or termination, to allocate tasks based on individual behaviour or traits, or to monitor and evaluate performance and behaviour (point 4(b)). A writing assistant that drafts offers or summarises meetings is not in that list.
The date matters too. The Annex III obligations were originally due on 2 August 2026. Regulation (EU) 2026/1744, in force since 27 July 2026, amended Article 113 and moved the date to 2 December 2027 for stand-alone Annex III systems, and to 2 August 2028 for AI embedded in regulated products under Annex I. We track the timeline in EU AI Act high-risk deadline delayed to 2027-2028.
Two provisions of the AI Act apply at the workplace already. Since 2 February 2025, Article 5(1)(f) prohibits AI systems that infer the emotions of workers at the workplace, except for medical or safety reasons; no notice or agreement cures that, see Prohibited AI practices under the EU AI Act. Since the same date, Article 4 requires employers to take measures so that staff using AI have sufficient AI literacy, which is what our training is built around.
Finally, Article 2(11) states that the Regulation does not preclude Member States from maintaining or introducing laws more favourable to workers, or from allowing more favourable collective agreements. National law stays fully in force.
GDPR: the layer that already applies everywhere
Independently of the AI Act, the GDPR has required transparency towards employees since 25 May 2018. Articles 13 and 14 oblige the employer to inform data subjects about every processing of their personal data, which includes an AI tool that stores prompts per user account or scores their output. Article 22 limits decisions based solely on automated processing that produce legal or similarly significant effects, a threshold that candidate screening and dismissal decisions reach quickly. Article 88 allows Member States to adopt more specific rules for processing in the employment context and names transparency, transfers within a group and monitoring systems at the workplace as areas those rules must address. Whether a data protection impact assessment is also due is covered in When does an AI system require a DPIA?.
Germany: the works council decides, not just listens
In Germany the works council (Betriebsrat) can be elected in establishments with at least five permanent employees entitled to vote (§ 1 BetrVG). The Works Council Modernisation Act of 14 June 2021 (BGBl. I p. 1762, in force since 18 June 2021) wrote artificial intelligence into four provisions.
§ 90(1) no. 3 BetrVG requires the employer to inform the works council in good time, with the necessary documents, about the planning of work processes including the use of artificial intelligence, and § 90(2) requires consultation early enough that the council's proposals can still influence the plan. § 87(1) no. 6 BetrVG goes further: the introduction and use of technical equipment designed to monitor the conduct or performance of employees is subject to co-determination, meaning the system cannot go live without a works agreement or a ruling by the conciliation committee. The Federal Labour Court reads "designed to monitor" as objective suitability; the employer's intention is irrelevant. It confirmed that line on 16 July 2024 (1 ABR 16/23) for a headset system in which supervisors could listen in live without any recording, as summarised by Gleiss Lutz. An AI assistant that logs prompts, timestamps and outputs per user meets that test, in our reading, without difficulty.
§ 80(3) BetrVG adds that when the works council has to assess the introduction or use of AI, calling in an expert is deemed necessary, at the employer's cost. And § 95(2a) BetrVG extends the council's consent right over selection guidelines for hiring, transfers and dismissals to cases where AI is used to draw them up.
Spain: the algorithm information right since 2021
Spain was the first Member State with a statutory right to information about algorithms at work. Royal Decree-Law 9/2021 of 11 May, published in the BOE on 12 May 2021 and in force since 12 August 2021 (later confirmed as Law 12/2021 of 28 September), added letter (d) to Article 64.4 of the Workers' Statute. The works committee has the right to be informed by the company of the parameters, rules and instructions on which the algorithms or artificial intelligence systems are based that affect decision-making with a possible impact on working conditions, access to and maintenance of employment, including profiling.
The right is not limited to delivery platforms, even though the law was triggered by them: it sits in the general catalogue of information rights of Article 64. It belongs to the works committee in workplaces with 50 or more employees (Article 63) and, under Article 62.2, equally to the staff delegates elected in workplaces with between 10 and 49 employees. Article 64.5 adds a right to a prior report on the introduction or revision of systems for organising and controlling work, which is where a productivity or scheduling tool lands. The Ministry of Labour presented a practical guide with a voluntary questionnaire on 10 June 2022 to help companies meet the duty; as law firm Cuatrecasas notes, it stops short of requiring disclosure of source code.
Platform work: the direction of travel
For digital labour platforms the EU has gone further. Directive (EU) 2024/2831 of 23 October 2024, published on 11 November 2024 and in force since 1 December 2024, must be transposed by 2 December 2026. Its Article 9 requires platforms to inform persons performing platform work and their representatives about automated monitoring and automated decision-making systems, at the latest on the first working day, before any change, and on request. The directive binds platforms only, but it shows the content that information duties are converging on: which systems exist, what they monitor, which decisions they take or support, and who oversees them.
Which rule applies when
| Rule | In force since | Trigger | Who must be informed or involved |
|---|---|---|---|
| GDPR Art. 13, 14, 22, 88 | 25 May 2018 | Processing of employee data | Each employee, information |
| Germany, § 90 BetrVG | 18 June 2021 (AI named) | Planning work processes with AI | Works council, information and consultation |
| Germany, § 87(1) no. 6 BetrVG | 1972, as read by the BAG | System objectively suited to monitoring | Works council, co-determination |
| Spain, Art. 64.4(d) Workers' Statute | 12 August 2021 | Algorithms affecting working conditions or employment | Works committee or staff delegates |
| AI Act Art. 5(1)(f) | 2 February 2025 | Emotion recognition at work | Prohibited, nothing to negotiate |
| AI Act Art. 4 | 2 February 2025 | Any AI use by staff | Staff, AI literacy |
| AI Act Art. 26(7) | 2 December 2027 | High-risk system under Annex III point 4 | Workers' representatives and affected workers |
| Directive 2024/2831 Art. 9 | 2 December 2026 (transposition) | Platforms using automated systems | Persons performing platform work |
What on-premise AI changes in the conversation
The information and co-determination duties do not depend on where the server stands. The content of the agreement does. In our experience, works agreements and committee consultations about AI stall on three questions: where employees' inputs go, who can read the logs, and whether the data is used for training. With a cloud service the employer can answer only as far as the provider's terms allow, and must reopen the discussion when those terms change.
With a model running on the company's own hardware, the same points can be fixed technically and verified by the representatives' expert: no transmission to third parties, logging limited to what the agreement permits, per-user evaluation disabled by configuration, no training on employee data because the model is not retrained. What such a setup contains is described on our local AI page. For Annex III use cases such as application screening, the full set of Article 26 deployer duties still arrives on 2 December 2027, from human oversight to log retention; an agreement signed now should already include the Article 26(7) notice so that no second round is needed.
A sensible order: inform the representatives before purchasing, with documents on purpose, data flows and logs; check whether the system is objectively suited to monitoring and, if so, negotiate the agreement before a pilot touches real users; update the GDPR privacy notice and schedule the Article 4 training in parallel; and if the use case is in Annex III point 4, put the December 2027 deadline into the same plan. A pilot project on local hardware is a good vehicle for this: the representatives see a running system, can inspect the logs themselves, and the agreement describes something concrete.
Frequently asked questions
Does the EU AI Act require employers to inform employees about AI?
Yes, for high-risk systems. Article 26(7) obliges employers to inform workers' representatives and the affected workers before putting a high-risk AI system into service at the workplace. Under Regulation (EU) 2026/1744 that duty applies from 2 December 2027 for the Annex III use cases, which include recruitment, task allocation and performance monitoring.
What applies today, before December 2027?
Three things. Article 5(1)(f) has banned emotion recognition at the workplace since 2 February 2025. Article 4 on AI literacy has applied since the same date. And national labour law already requires involvement of worker representatives: in Germany through the works council under the BetrVG, in Spain through the information right on algorithms in Article 64.4(d) of the Workers' Statute, in force since 12 August 2021.
Does GDPR also require informing employees about AI tools?
Yes. Articles 13 and 14 require information about any processing of employee data, including an AI tool that logs or evaluates their input. Article 22 restricts decisions based solely on automated processing, and Article 88 lets Member States set stricter rules for the employment context, explicitly naming transparency and monitoring systems.
Does running AI on-premise change the information duties?
The duties stay the same, the conversation changes. With a local system the employer can commit that no input leaves the company, that logs stay in house and that per-user evaluation is switched off by configuration. Those are the points on which negotiations with worker representatives about cloud tools usually stall.