ChatGPT at work: no co-determination for private accounts

Approval without the works council

A global manufacturer of medical technology employs around 1,600 people at a joint operation in Hamburg. In December 2023, the company briefly blocked access to ChatGPT on its systems and unblocked it again on 18 December. On the same day, two guidelines on the use of generative AI and a manual appeared on the intranet. Alongside them came a statement to the workforce encouraging the use of generative AI as a new tool for work.

The technical side carries the entire decision. ChatGPT and comparable systems were not installed on the company's computers. Anyone who wanted to use them opened them in a browser and had to set up their own private account with the provider. The company did not issue work accounts. Where costs arose, employees bore them themselves. The company knew neither who had an account nor when, for how long or for what purpose someone used the tool, and use was voluntary.

The group works council, that is, the works council at group level, demanded the very same day that access be blocked again until a framework agreement on AI was in place. The company definitively refused in early January 2024. The works council brought urgent proceedings before the Hamburg Labour Court. With eleven applications, it sought to have the guidelines removed from the intranet and the use of ChatGPT prohibited until it gave its consent or the conciliation committee, as the arbitration body, decided. By order of 16 January 2024 (case no. 24 BVGa 1/24), the court dismissed all the applications; the order is final.

Working tool rather than workplace conduct

The works council relied first on its co-determination right concerning the rules of operation. According to the settled case law of the Federal Labour Court, this right applies only to so-called conduct-related behaviour, that is, rules on how the workforce lives and works together, such as a dress code. Work-related behaviour, by contrast, is free of co-determination: by virtue of its right to issue instructions, the employer determines what work is to be done and in what manner.

The Labour Court classified the AI guidelines as work-related behaviour. The company was making a new working tool available to its employees under certain conditions. The guideline and the manual were instructions on how the work was to be performed. There was therefore no co-determination right.

The court did not accept the works council's argument that voluntary use split the workforce into those open to AI and those sceptical of it. Otherwise every new working tool that was not immediately given to everyone would trigger co-determination, even though the employer is merely directing how work is done.

No access, no monitoring

More important in practice is the second point: co-determination in relation to technical devices designed to monitor the conduct or performance of employees. Monitoring in this sense means that the employer uses technical means to collect and store information about conduct or performance and can access it later. The device must collect the data itself. Whether the employer intends to monitor is irrelevant; it is enough that the technology is objectively suitable for doing so.

That was lacking here. ChatGPT was not installed. Employees opened it in the browser like any other website. The browser does store the history and is therefore itself a technical device capable of recording behavioural data. But a group works agreement on browser use was already in place. The works council had therefore already exercised this co-determination right.

As for the tool itself, the court saw no monitoring pressure emanating from the employer. Because employees set up their own accounts and bear the costs, the company receives no report of who uses ChatGPT, when, for how long and for what purpose. The court assumed that the provider records such data. But the company cannot access it. Nor did the obligation to label AI-assisted work results change anything. That check is performed by the employee through their labelling; the tool collects nothing in the process.

Health and safety and information

Two further arguments raised by the works council were also unsuccessful. Co-determination in health and safety matters requires a specific hazard that either exists or has been identified in a risk assessment. The works council had referred only in general terms to possible psychological strain caused by new software, and that was not enough for the court.

The chamber left open whether the company had informed and consulted the works council in good time, as the law requires for new working methods. The company considered this duty fulfilled. It did not matter for the urgent application: information and consultation give the works council no right of veto, and a single breach would not be a gross breach of duty justifying a court injunction. Several applications also failed because of their breadth. A prohibition on all "applications involving generative artificial intelligence" was too vague for the court.

Where the works council has a say on ChatGPT

The reasoning holds only as long as the circumstances remain the same: no installation on company systems, private accounts, costs borne by employees, no usage data held by the employer, voluntary use and an existing agreement on the browser. The court itself noted that work accounts are "currently" not being issued. The company argued that access to the data was technically impossible as long as there were no company accounts. Court and employer therefore assumed that company accounts would change the position.

Commentators accordingly read the order narrowly. Where a company introduces AI software itself, for example with company licences or on its own systems, and can see who uses the tool and how, co-determination is a real prospect. A duty to use the tool is also cited as a circumstance capable of changing the outcome. The Hamburg Labour Court did not decide these points, and this was a single chamber in urgent proceedings.

For companies, that leaves a clear but narrow path. On this decision, anyone who permits employees to use ChatGPT via private accounts and lays down rules for it does not need the works council's consent. The rules may specify how the tool is to be used at work and that AI results must be labelled. The duty to inform and consult the works council in good time remains unaffected. And the decision says nothing about data protection. The company stated that it had previously carried out a data protection impact assessment, that is, a documented risk review. The order relieves no company of that review.