Vademecum of the GPAI Signatory Taskforce

1 VADEMECUM THE SIGNATORY TASKFORCE OF THE GENERAL-PURPOSE AI CODE OF PRACTICE January 2026 Having regard to the General -Purpose AI Code of Practice published on 10 July 2025 ( the ‘Code’), assessed as adequate pursuant to Article 56 (6) AI Act, this Vademecum sets out the organisation and operation of the Signatory Taskforce (the ‘Taskforce’), established by consensus of the attached list of Signatories to the Code (the ‘Signatories’) and agreed by the European AI Office (the ‘AI Office’). Point 1 Composition and membership of the Taskforce 1. The Taskforce is comprised of the attached list of Signatories as ‘Members’. The Taskforce is chaired by the AI Office. 2. Signatories that are not Members may communicate at any time their intent to join the Taskforce to the AI Office. If these Signatories have general-purpose AI models, they shall become Members and the AI Office shall include them in the next Taskforce meeting. 3. Members may communicate at any time their intent to withdraw from the Taskforce to the AI Office. The AI Office shall work with the withdrawing Member to promptly remove them from meetings moving forward. Point 2 Objectives The objectives of the Taskforce are, inter alia: (a) to facilitate cooperation between Signatories where relevant for the purpos e of implementing the Code; (b) to provide input on guidance documents produced by the AI Office that are relevant to the Code , in view of technological , societal, market , and legislative developments, without prejudice to the necessary public consultations by the AI Office on any such possible adaptations or documents; (c) to exchange views on technological developments, changes in the risk landscape, or experience with the application of the AI Act rules that could be relevant for potential future reviews or updates of the Code , without prejudice to the necessary public consultations by the AI Office on any such possible updates; (d) to gather insights from, organise exchanges with, and provide updates to relevant third- party stakeholders on developments in the field of general-purpose AI models for the purpose of implementing the Code; and 2 (e) where appropriate, seeking and discussing research, independent expert input, and up- to-date evidence relevant to the Code’s commitments or measures, such as emerging practices. Point 3 Ad hoc groups 1. The Taskforce may set up ad hoc groups as necessary to examine specific issues , in line with the decision-making procedure under Point 5 . These ad hoc groups will be made up of the Members whose activities, expertise, or resources are relevant to the purposes of the group in question. 2. The rules for the Taskforce set out in this Vademecum apply to the se ad hoc groups mutatis mutandis. Point 4 Meetings 1. The Taskforce shall meet at a frequency as required to carry out the Taskforce’s objectives and in any event at least yearly. Meetings of the Taskforce shall be convened by the AI Office either on its own initiative or at the request of a simple majority of the Members. The AI Office, as chair, shall facilitate discussions. 2. Meetings can be conducted in a hybrid format, i.e., in person with the possibility of joining remotely, or online. The AI Office shall provide logistical support to the Taskforce. Following the consultation with Members, the AI Office shall propose agendas for the meetings of the Taskforce and circulate them to the Members, and to invited experts and stakeholders (Point 6) as appropriate, at least two weeks in advance of the meetings. The AI Office shall also draw up high-level meeting notes for circulation to the Members. 3. The Members shall comply with applicable Union law, including in particular competition rules, for all their activities, meetings, and exchanges as part of the Taskforce. Point 5 Decision-making procedure 1. The Taskforce may adopt outputs relating to its Objectives under Point 2 by consensus based on all Members . Agreement shall be assumed, u nless a Member expresses its disagreement , within a reasonable period of time set by the AI Office, which shall be no less than 10 working days. Members may append individual statements to an output. 2. Paragraph 1 shall also apply to any amendments to this Vademecum, with agreement by the AI Office. 3 Point 6 Invited experts and stakeholders The Taskforce may — upon the proposal of one or more Members or the AI Office , and supported by a two-thirds majority of the Members (Point 5 applies mutatis mutandis) — invite experts or stakeholders to present their views on technical matters or specific aspects of the Code. In the invite, such experts or stakeholders may be permitted by the Taskforce to attend specific meetings or agenda items of the Taskforce related to their input on a case-by-case basis. They can take part in discussions, but they cannot vote on outputs developed by the Taskforce. Point 7 Confidentiality 1. To ensure open and sincere discussions, the Members and invited experts or stakeholders must keep the discussions, drafts , and outputs of the Taskforce confidential vis-à-vis third parties , unless the Taskforce agrees otherwise . This includes any press outreach or public communication on the ongoing work. Unless otherwise agreed by the Taskforce, discussions held in the Taskforce are off the record. Such agreements require consensus in line with t he decision-making procedure under Point 5. 2. The AI Office may publish this Vademecum, the list of organisations participating in any meeting, and the high-level minutes of meetings as required by its transparency obligations. 4 LIST OF PARTICIPATING SIGNATORIES IN THE TASKFORCE Accexible AI Alignment Solutions Aleph Alpha Amazon Anthropic Black Forest Labs Bria Cohere Cyber Institute Domyn Dweve Fastweb Google Humane Technology IBM LINAGORA Microsoft Mistral OpenAI Open Hippo ServiceNow Virtuo Turing Writer