⚡ Quick Summary
This document is the Transparency Chapter of the Code of Practice for General-Purpose AI Models, introduced by Nuria Oliver, Working Group 1 Co-Chair, and Rishi Bommasani, Working Group 1 Vice-Chair. It describes three Measures that Signatories commit to implementing in order to comply with their transparency obligations under Article 53(1), points (a) and (b), and the corresponding Annexes XI and XII of the AI Act.
The Chapter opens with an introductory note, objectives and recitals, and then sets out Commitment 1 Documentation. Measure 1.1 requires signatories to document at least the information in the Model Documentation Form, keep it up to date and retain previous versions for ten years after the model is placed on the market. Measure 1.2 covers disclosure: publishing contact information, answering AI Office requests within the period specified under Article 91(4), and giving downstream providers information no later than 14 days after a request. Measure 1.3 requires quality, integrity and security of the documented information. The Model Documentation Form indicates for each item whether it is intended for downstream providers, the AI Office or national competent authorities.
🧩 What’s Covered
The Chapter follows the structure of a code-of-practice commitment, with explanatory sections followed by measures and a documentation form.
- Introductory note: the Chapter addresses transparency obligations under Article 53(1), points (a) and (b), and Annexes XI and XII; it introduces the Model Documentation Form that lets signatories compile the required information in one place, states that items intended for the AI Office or national competent authorities are released only on request, and records the Article 78 confidentiality obligation.
- Objectives: the overarching objective drawn from Article 1(1) AI Act and two specific objectives, namely to serve as a guiding document for demonstrating compliance with Articles 53 and 55 while noting that adherence is not conclusive evidence of compliance, and to enable the AI Office to assess compliance.
- Recitals (a)–(c): the role of general-purpose AI model providers in the value chain (recital 101), proportionality when a model is fine-tuned or otherwise modified (recital 109), and the need to account for market and technological developments (Article 56(2), point (a)).
- Commitment 1 Documentation and Measures 1.1–1.3: the legal text cited; drawing up and updating the Model Documentation; providing information to the AI Office on request and to downstream providers within 14 days; and ensuring quality, integrity and security of the information.
- Article 53(2) exception: the Measures do not apply to models released under a free and open-source licence meeting the conditions in that provision, unless the model has systemic risk.
- Deferral: documentation under Annex XI, Section 2, applicable to models with systemic risk, is covered by Measure 10.1 of the Safety and Security Chapter.
- Model Documentation Form: presented as a static version whose input fields cannot be filled in, with a fillable version separately available.
💡 Why it matters?
For providers of general-purpose AI models, the Chapter turns the documentation duty in Article 53(1), points (a) and (b), into concrete measures: what to document, how long to retain it, how quickly to answer requests and how to protect the information. Downstream providers can see what they may request and within what period, which matters where a high-risk AI system is built on a model supplied by another provider. The split between information intended for downstream providers and information released only to the AI Office or national competent authorities, together with the Article 78 confidentiality safeguards, is the practical core for legal and compliance teams.
❓ What’s Missing
The static Model Documentation Form is a placeholder: its fields, which carry the substance of what must be documented, are not reproduced in the extracted text, and the fillable version is only said to exist separately. No publisher, publication date or version number is printed, so the state of the Code cannot be verified from the document itself. Documentation for models with systemic risk is deferred to Measure 10.1 of the Safety and Security Chapter and is not covered here. Enforcement beyond AI Office requests, penalties and worked examples of completed documentation are absent.
👥 Best For
Compliance and legal teams at providers of general-purpose AI models who must evidence Article 53(1) documentation; downstream providers and integrators working out what model information they can request and within what timeframe; AI Office and national competent authority staff exercising supervisory tasks; policy analysts tracking the AI Act's code-of-practice instruments.
📄 Source Details
The document is the Transparency Chapter of the Code of Practice for General-Purpose AI Models. No publisher, publication year, version or reference number is printed in the extracted text. Nuria Oliver is named on the cover as Working Group 1 Co-Chair and Rishi Bommasani as Working Group 1 Vice-Chair; no author byline is used elsewhere. The text runs to six pages in English. The extraction covered all six pages, ending with the note that the Model Documentation Form shown is a static version whose input fields cannot be filled in.
