Artificial intelligence is making its way into companies a little at a time: a writing assistant here, access to a generative AI tool there, an “innovation” working group somewhere else. Seen from management, nothing that looks like a major project. That is precisely where the risk lies.
In a judgment of 21 May 2026[1], the Paris Court of Appeal made the point plainly: the introduction of AI tools may constitute a “new technology” subject to prior information and consultation of the works council (comité social et économique, CSE), and deploying them without consultation may simply be suspended.
AI is therefore no longer merely a matter of innovation or IT: it has become an employment law issue.
For employers, the aim is not to freeze innovation, but to identify the point at which making a tool available ceases to be a technical choice and becomes a decision subject to social dialogue.
I. The introduction of AI, a new technology within the meaning of employment law
In companies with at least fifty employees, the works council must be informed and consulted on the general running of the company, and in particular on the introduction of new technologies or any significant change affecting working conditions[2].
Two systems were at issue: an in-house writing assistant, capable of transcribing a recording, rephrasing a text or suggesting headlines, and access, governed by a charter, to an external general-purpose generative AI tool.
For the Court, there was no doubt: these tools are new technologies, liable to have a significant effect on working conditions and on the nature and volume of the tasks assigned, first and foremost those of journalists and sub-editors.
Consultation is not, however, automatic as soon as a piece of software includes an AI function. The assessment remains a concrete one: the significance of the effects on work, the number of employees concerned and the lasting nature of the project.
It is precisely this case-by-case approach that makes the subject a delicate one for employers.
II. The optional nature of the tool is not enough to rule out the obligation to consult
The employer put forward the argument heard everywhere: the tool is optional, and some employees were already using generative AI tools on their own initiative.
The Court rejected it.
Making a tool available on the intranet encourages its adoption, regardless of the fact that it is not compulsory. As for spontaneous use (informal, or even prohibited for data protection reasons), it does not amount to a decision by the employer.
The distinction is essential: it is the decision to organise, authorise or regulate the use of AI that triggers social dialogue, not an individual employee’s isolated click. In other words, regulating a use already amounts to introducing it.
III. Heightened vigilance for evolving projects
AI is rarely deployed all at once: experimentation, restricted access, general roll-out, integration into business tools, extension of functionalities, and then formalisation of a usage policy.
The gradual nature of the project neither allows consultation to be postponed indefinitely nor reduced to a general, abstract presentation. Where the roll-out forms part of a complex procedure with staggered decisions, the council must be consulted at each significant stage.
The nuance remains: a mere technological tool, with no demonstrated effect on working conditions, is not enough to constitute a significant project. But as soon as AI transforms methods, tasks, the skills required or the monitoring of work, the risk becomes real.
IV. An immediate litigation risk for the employer
The sanction is anything but theoretical. Without prior consultation, the works council applies to the urgent applications judge (juge des référés), who may suspend the roll-out until the end of the procedure and attach a periodic penalty payment (astreinte) to the injunction[3].
In the case decided, the delay of more than two months in opening the consultation justified the periodic penalty payment.
And the trend goes beyond a single decision: other courts have taken the same path, notably in relation to a platform giving access to generative AI tools[4].
For the employer, the calculation is simple: consultation costs a few weeks; overlooking it costs the halting of the project at the worst possible moment, damages and a lawsuit.
Hence a method. Before any roll-out, map the intended uses, the employees concerned and the foreseeable effects on organisation, the skills required and the monitoring of work, as well as on health, safety and data. It is this analysis that will show whether consultation is required, and at what stage of the project.
And a safeguard. A charter on the use of digital tools setting out the authorised uses, their limits, confidentiality rules, the requirement for human oversight and the precautions applicable to data.
It has the merit of also addressing spontaneous uses, which often precede any formal decision by the employer. Beyond the works council risk alone, the charter is the simplest tool for securing the day-to-day use of AI: it turns an unmanaged practice into a controlled framework, and applies as much to in-house tools as to consumer generative AI.
AI is a remarkable lever for efficiency. But introducing it is not a mere technical choice: it is a decision that requires legal preparation, social dialogue included
[1]CA Paris, pôle 1 ch. 2, 21 May 2026, No. 25/13232
[2]C. trav., Art. L. 2312-8, II, 4°; consultation must take place before the employer’s decision: Art. L. 2312-14.
[3]Principle restated by Cass. soc., 27 November 2024, No. 23-13.806, published in the Bulletin.
[4]TJ Paris, ord. réf., 2 September 2025.
