For employers in the US and Europe introducing AI tools or agents, this guide explains when employees or their representatives must be told, consulted or asked to agree, and what to do before a tool goes live.
What you'll learn
- What US labor law requires of union and non-union employers
- The EU's information and consultation rules and the AI Act's duty to inform workers
- When a German works council has a say, how the Netherlands and the UK compare, and the first court decision on ChatGPT
- Which AI tools count as high-risk for workers under the AI Act
- The steps before go-live, and what goes into a workplace AI agreement
Whether employees get a say when you introduce AI depends on where you operate and what the tool can record about them. In the US, the duty runs through union bargaining, and only 5.9% of private-sector workers were union members in 2025. In the EU, employers with 50 or more staff must inform and consult employee representatives on substantial changes to work organization, and from December 2, 2027 the AI Act requires them to tell workers before using high-risk AI on them. Germany goes furthest: a works council has a say whenever a tool is designed to monitor behavior or performance. This guide covers what applies, which tools trigger it, and the steps before go-live.
Most AI rollouts are planned by IT and management and announced to everyone else at launch. In some places that is lawful. In others it can stop the rollout, because employees or their representatives had a right to be heard first.
This guide sets out when that right exists in the US, in the EU, and in Germany, where it is strongest. It also covers which tools trigger it, and what to do before go-live so the employee step costs weeks rather than months.
Do employees get a say when you introduce AI at work?
Sometimes, depending on where you operate and what the tool can record about employees. In the US, the right runs through union bargaining, and only 5.9% of private-sector workers were union members in 2025.[1][2] In the EU, employers must inform and consult on substantial changes to work organization, and the AI Act adds a duty to inform workers about high-risk AI from December 2, 2027.[3][4][6] Germany adds the strongest right: co-determination over tools that monitor employees.[8]
| Where | Who has a say | When it applies |
|---|---|---|
| US | A union, where employees have chosen one | Bargaining over wages, hours and other terms and conditions of employment[1] |
| EU | Employee representatives | Information and consultation on substantial changes to work organization, from 50 employees (or 20 per establishment)[3] |
| EU, AI Act | Workers' representatives and affected workers | Information before using high-risk AI on workers, from December 2, 2027[4][6] |
| Germany | The works council | Information on all AI planning; agreement for tools designed to monitor behavior or performance[7][8] |
The rows build on each other. A company in Germany has to meet the EU rules as well as its own, and a US company with staff in Europe meets the European rules for those staff.
What does US law require?
US law gives employees a say through their union, where they have one. Section 8(d) of the National Labor Relations Act obliges employers to "confer in good faith with respect to wages, hours, and other terms and conditions of employment" with the employees' representative.[1] AI that changes how work is done, measured or paid can fall within those terms. In 2025, 10.0% of US wage and salary workers were union members, and 5.9% in the private sector.[2]
For the roughly 94% of private-sector workers who are not union members, there is usually no union to bargain with. Other sources of employee rights are union contracts, which can contain their own technology clauses, and from state and city law. New York City, for example, has required since July 2023 that automated tools used to screen candidates or employees for hiring or promotion pass a yearly bias audit, with notices to the people assessed.[12] Check the states and cities where your employees work before an AI tool starts deciding or scoring anything about them.
Where a union represents your employees, raise AI early. Introducing a tool that changes workloads or performance measures without bargaining invites a dispute; agreeing the rules first rarely costs more than a few meetings.
What does EU law require?
EU law requires employers to inform and consult employee representatives before substantial changes, and adds an AI-specific duty from December 2027. Directive 2002/14/EC covers decisions "likely to lead to substantial changes in work organisation", in undertakings with at least 50 employees or establishments with at least 20, depending on the member state.[3] Information must come early enough for representatives "to conduct an adequate study".[3]
The AI Act adds a separate duty. Before putting a high-risk AI system into use at the workplace, employers "shall inform workers' representatives and the affected workers".[4] The Digital Omnibus moved the date: for high-risk systems listed in Annex III, which include most workplace uses, the rules apply from December 2, 2027.[6]
Each member state implements the directive in its own way, through works councils, staff committees or other bodies. The thresholds and procedures differ, so check the rules in each country where you employ people. Germany, covered next, is the strictest common case.
When does a German works council have a say?
A German works council must be informed about planned AI use in every case, and has a co-determination right when a tool is designed to monitor employees. Section 90 of the Works Constitution Act (BetrVG) names "the use of artificial intelligence" among the work processes the employer must inform the council about in good time, with the documents, and consult on.[7] Section 87(1) no. 6 gives the council a say on technical systems designed to monitor behavior or performance.[8]
Where the two sides do not agree on a monitoring tool, a conciliation committee decides.[8] When a works council has to assess AI, the law treats calling in an outside expert as necessary, by arrangement with the employer.[9] In 2024, about 7% of private establishments with five or more employees had a works council, but those are the larger ones: about 37% of employees worked in one.[11]
The practical consequence is timing. Section 90 requires information at the planning stage, so that the council's proposals and concerns can still be taken into account.[7] An employer that presents a finished, configured AI tool at launch has usually missed that point.
How do the Netherlands and the UK compare?
The Netherlands is close to Germany; the UK is closer to the US. Under article 27(1)(l) of the Dutch Works Councils Act (WOR), an employer needs the works council's consent for any arrangement on systems aimed at, or suitable for, observing or checking employees' presence, behavior or performance.[13] Dutch businesses with at least 50 people must have a works council, and article 27(1)(k) adds consent for rules on processing employees' personal data.[13]
The Dutch wording matters for AI. A tool only has to be suitable for monitoring, not intended for it, so an agent that logs who asked what can need consent even if nobody plans to look at the logs.[13]
The UK has no standing works council by default. Under the Information and Consultation of Employees Regulations 2004, employees of undertakings with at least 50 staff can ask for an information and consultation agreement; since April 2020, a request from 2% of employees, with a minimum of 15, is enough.[14] Where no request has been made and no union is recognized, the regulations do not themselves require consultation.[14]
Does allowing ChatGPT need works council approval?
Not necessarily. In January 2024, the Hamburg Labor Court rejected a group works council's application to stop employees using ChatGPT, in interim proceedings.[10] The employer had allowed and encouraged use through the browser and employees' private accounts, published usage guidelines, and installed nothing; it received no data about how employees used the tool, so the court saw no monitoring.[10]
The decision is narrower than it sounds. Law firm Gleiss Lutz, reporting on it, warned that small differences in the facts could change the result, especially where an employer runs its own AI systems or has access to the data they record.[10] Company accounts with usage logs, an agent that records which employee asked what, or dashboards of AI use per person look different.
The question is rarely whether a tool uses AI. It is whether the tool can watch the people who use it.
Most business AI tools record something: who logged in, what they asked, which tasks they completed. Our guide on what an AI agent can see covers how to decide what an agent records and who gets access to it. Those decisions are exactly what a works council will ask about.
Which AI tools count as high-risk for workers?
The AI Act treats workplace AI as high-risk when it makes or shapes decisions about people. Annex III lists AI used for recruitment and selection, and AI used "to make decisions affecting terms of work-related relationships", promotion or termination, task allocation based on individual behavior or personal traits, and monitoring or evaluating performance and behavior.[5] From December 2, 2027, using such a system triggers the duty to inform workers first.[4][6]
Most AI tools in daily office use are not on that list. A writing assistant, a meeting summarizer or an agent that drafts quotes does not decide anything about the employee using it. The same agent becomes high-risk when its output is used to rank employees, assign work by individual traits or evaluate performance.[5]
That makes the purpose you configure decisive. Decide in writing what each AI tool is for and what it is not used for, and keep the second list as concrete as the first. Our EU AI Act guide for small businesses covers the other duties of companies that use AI.
What should you do before an AI tool goes live?
Bring employees in at the planning stage, before the tool is configured. In Germany that is a legal requirement for all AI planning, and in the EU the consultation duty also requires information early enough for a proper review.[3][7] These six steps work under all three systems and turn the employee step into a few weeks of planning rather than a stopped rollout.
Map who the tool touches
List what it records
Decide what it is not for
Inform representatives early
Agree the rules
Tell the employees
What goes into a workplace AI agreement?
A workplace AI agreement sets out what a tool is for, what it records and what may never be done with the records. In Germany such agreements usually take the form of a works agreement (Betriebsvereinbarung) under the co-determination rules; elsewhere the same points work as a policy agreed with the union or published to staff.[8] The following points cover what representatives typically ask about first.
| Topic | What the agreement settles |
|---|---|
| Purpose | Which tasks the tool is for, and which uses are excluded |
| Tools covered | Named tools and versions, and how new ones are added |
| Data recorded | What is logged about employees, where, for how long |
| Access to records | Who may see logs and for what reasons |
| Performance use | Whether records may be used to evaluate individuals (usually: no) |
| Training | What employees learn before using the tool |
| Review | When the agreement is checked again, for example yearly |
An agreement that names the tool, the records and their limits is quicker to approve than a general clause about "AI". It also protects the employer: it shows employees and authorities what the tool does, in writing, before anyone has a complaint.
When does this matter for an AI agent?
The employee question matters as soon as an agent works alongside staff or acts on their work: answering their questions, routing their tasks, or reviewing what they produce. An agent that records who asked what can monitor employees whether or not anyone intends it, which in Germany brings the works council in.[8] An agent built with its records and access decided up front makes that conversation short.
The rollout itself also needs a plan for the people using the agent. Our guide on how to onboard an AI agent covers the first 30 days. For the choice between a general AI tool and an agent built around your own processes and approval steps, see what a custom agent is and when to build one.
Frequently asked questions
Do you have to tell employees before introducing AI at work? It depends on where you operate. In the US, unionized employers must bargain over changes to terms and conditions of employment, and some states and cities add rules, such as New York City's bias-audit and notice law for automated hiring and promotion tools. In the EU, employers with 50 or more staff must inform and consult employee representatives on substantial changes to work organization, and from December 2, 2027 the AI Act requires employers to inform workers before using high-risk AI on them. Germany requires informing the works council about planned AI use in every case.
Does a works council have to approve AI tools? In Germany, a works council has a co-determination right when a tool is designed to monitor employees' behavior or performance, and without agreement a conciliation committee decides. For other AI tools it must be informed and consulted in good time but cannot block them. A Hamburg labor court ruled in January 2024 that letting staff use ChatGPT through private accounts, with no employee data reaching the employer, needed no works council agreement.
What AI systems count as high-risk in the workplace under the EU AI Act? Annex III of the AI Act lists AI used to recruit or select people, to make decisions on promotion, termination or terms of work, to allocate tasks based on individual behavior or personal traits, and to monitor or evaluate workers' performance and behavior. Since the Digital Omnibus, the obligations for these systems apply from December 2, 2027. Employers must inform workers' representatives and affected workers before using them.
Can employees refuse to use AI at work? The law on this differs by country and contract, and no general rule applies across the US or the EU. In practice, the question is easier when employees know what the tool does, what it records about them and who sees it. Telling employees early, and agreeing rules with their representatives where they exist, avoids most of the conflict.
Notes
- National Labor Relations Act, section 8(d), 29 U.S. Code § 158, Legal Information Institute, accessed October 16, 2026.
- U.S. Bureau of Labor Statistics, Union Members — 2025, news release USDL-26-0229, February 18, 2026, accessed October 16, 2026.
- Directive 2002/14/EC establishing a general framework for informing and consulting employees, Articles 3 and 4, accessed October 16, 2026.
- AI Act Service Desk, Article 26: Obligations of deployers of high-risk AI systems, paragraph 7, accessed October 16, 2026.
- AI Act Service Desk, Annex III: High-risk AI systems, point 4, accessed October 16, 2026.
- AI Act Service Desk, Article 113: Entry into force and application, consolidated text as at July 27, 2026, accessed October 16, 2026.
- Works Constitution Act (BetrVG), section 90, in German, accessed October 16, 2026.
- Works Constitution Act (BetrVG), section 87, in German, accessed October 16, 2026.
- Works Constitution Act (BetrVG), section 80, in German, accessed October 16, 2026.
- Gleiss Lutz, Arbeitsgericht Hamburg zu ChatGPT: Kein Mitbestimmungsrecht des Betriebsrats, March 27, 2024, on Arbeitsgericht Hamburg, decision of January 16, 2024, 24 BVGa 1/24, in German, accessed October 16, 2026.
- Hohendanner and Kohaut, Tarifbindung und betriebliche Interessenvertretung: Ergebnisse aus dem IAB-Betriebspanel 2024, WSI-Mitteilungen 78 (4/2025), in German, accessed October 16, 2026.
- New York City Department of Consumer and Worker Protection, Automated Employment Decision Tools (AEDT), Local Law 144 of 2021, enforced from July 5, 2023, accessed October 16, 2026.
- Wet op de ondernemingsraden (Dutch Works Councils Act), articles 2 and 27, in Dutch, accessed October 16, 2026.
- The Information and Consultation of Employees Regulations 2004, regulation 7, and Schedule 1, legislation.gov.uk, accessed October 16, 2026.
