On 30 September 2026, the CFE-CGC Groupe RATP, a union representing management staff, sent a social alert to Jean Agulhon, the group's human resources director. The one-page letter, published on the union's website and signed by Valérie Vovk, central union delegate and vice-president of the organization, cited a single reason: "Non-respect of social dialogue regarding the deployment of AI in the company and non-respect of employee representative bodies in terms of information and consultation on projects for introducing new technologies in the company." The letter does not name any specific tool or project, nor does it provide figures or make requests beyond meeting with management representatives. The union claims that the management has not respected social dialogue and has failed to consult employee representative bodies on information and consultation regarding new technologies. According to Article L. 2312-8 of the labor code, the social and economic committee (CSE) must be informed and consulted on issues concerning the organization and operation of the company, including the introduction of new technologies and major changes affecting health and safety conditions or working conditions. These rules apply to the RATP, as stated in Article L. 2311-1, which states they are applicable to public establishments with industrial and commercial character. In previous cases, courts have ruled in favor of committees that felt bypassed. On 14 February 2025, the Nanterre judicial court ordered a company to suspend the deployment of five information tools until the completion of the consultation of the CSE initiated on 26 September 2024. Similarly, on 15 July 2025, the Créteil judicial court suspended the use of AI tools in the press sector until the consultation was closed. On 21 May 2026, the Paris court of appeal confirmed this order. The company involved was Groupe Industrie Services Info (GISI), a subsidiary of Infopro Digital, its editorial assistant DIGI, and the authorized access to ChatGPT. However, not all courts have followed the elected officials. On 10 February 2026, the Paris judicial court ruled that Copilot 365 is not a significant project requiring the consultation of the CSE. In a note dated 13 April 2026, the same authors mentioned an order from the Nanterre judicial court of 29 January 2026, which required a consultation for the replacement of a human resources software already incorporating AI. In a leaflet dated June 2026, the CFE-CGC Groupe RATP indicated that it had been received in audience by the management after having been alerted "several times" on "the absence of social dialogue" around AI. It listed nine demands, including the absence of job elimination linked to AI, a register of AI systems, an impact study before each deployment, and the sharing of productivity gains. The same document reports that the management proposed to open the negotiation of a method agreement in the last quarter of 2026, and that the union responded that, in the meantime, any deployment or experimentation introducing AI must be subject to information-consultation. According to La Revue du Digital on 1 October, Fabien Renaud and Valérie Vovk, president and vice-president of the CFE-CGC Groupe RATP, demand that the deployment be done "in transparency, with clear governance, a real evaluation of its impacts, and a consultation with the personnel representatives." The union states that it is not opposed to AI and notes that the management proposes to use it notably in human resources, which it considers "antithetical" to "human management of resources." The social alert is a procedure that the RATP collective agreement presents as a homegrown creation. Article 70 of the annex No. 3 to the agreement on the quality of social dialogue, signed on 7 July 2022, recalls that the RATP "was the originator of the social alarm prevention device against conflicts," adopted by the law of 21 August 2007 on the continuity of public service in transport. The transport code, in article L. 1324-2, has provided since that a strike notice cannot intervene "only after a prior negotiation" with the unions that intend to submit it. Only union delegates can trigger the alarm, with one reason per alarm. The management must then "hold a meeting with the alarm authors and draft the agreement or disagreement report within five working days" following receipt, unless an additional period is agreed upon. For a letter received on 30 September, this deadline expires on 6 October. The RATP professional elections take place from 19 to 23 October 2026, according to the union, which makes AI a campaign theme: "For an AI serving employees, and not the reverse, from 19 to 23 October, vote CFE-CGC Groupe RATP," concludes a text it published on 25 September. The alarm comes from a single union, that of management, and was submitted less than three weeks before the vote. This context says nothing about the validity of the grievance, but highlights the choice to make it public. The European AI regulation also requires informing the elected officials. Its Article 26, paragraph 7, obliges the employer who puts into service a high-risk AI system at the workplace to inform the representatives of the workers and the affected employees in advance. For the systems in Annex III, whose point 4 refers to the use and management of workers, Article 113 modified by Regulation (EU) 2026/1744 postpones the application to 2 December 2027. The tools described in the documents cited above (maintenance, graffiti detection, business assistant) do not concern employment decisions; the HR uses mentioned by the union are not detailed publicly. In France, the obligation to consult the CSE applies without waiting for this date. For an employer, the legal dispute revolves around two questions: does the tool affect working conditions, and has it been implemented before the CSE's opinion? A pilot expanded before this opinion exposes to a reference, which is resolved on documents: dates of information, documents transmitted, opinion rendered or not.