The Code was developed by independent experts in collaboration with over 1,400 stakeholders and is organised into three chapters: Transparency, Copyright, and Safety & Security. It includes a requirement for AI providers to ‘put in place a policy to comply with EU copyright law’ (Article 53(1c)), which is of particular relevance to the cultural sector given the use of cultural content in AI training without authorisation or remuneration.
General-purpose AI providers operating in the EU are encouraged to sign the Code. While non-binding, adherence can demonstrate alignment with the AI Act, potentially reducing administrative requirements. Non-signatories may face greater regulatory scrutiny.
Measure 1.1 – Copyright policy
Signatories must develop, maintain, and implement a copyright policy based on the Code’s measures, with designated organisational responsibility for oversight. Publication of a summary is encouraged.
Measure 1.2 – Lawful data use
Signatories commit to using only lawfully accessible, copyright-protected material when collecting data for AI training. They may not bypass paywalls, subscription models, or use pirated content. Piracy websites are defined as those infringing copyright ‘on a commercial scale’, with the European Commission set to publish a dynamic list of such sites.
Measure 1.3 – Respecting opt-outs
When crawling the internet for training data, signatories must respect machine-readable opt-outs, including the robots.txt protocol and any future versions. They must also recognise other machine-readable opt-out tools if officially adopted by EU or international standardisation bodies, or if they are widely used and endorsed at EU level. Signatories are required to publish information on their crawlers and notify rightsholders of updates.
Measure 1.4 – Preventing infringement by AI outputs
Signatories must take measures to prevent their AI models from producing copyright-infringing content and must include prohibitions on such uses in their acceptable use policies.
Measure 1.5 – Complaint mechanism
Signatories commit to designating an electronic point of contact for rightsholders to submit copyright-related complaints. Complaints must be handled diligently and within a reasonable time frame, unless manifestly unfounded or duplicative.
Changes from earlier versions
In the final version, most copyright measures have been elevated from ‘best’ or ‘reasonable’ efforts to firm commitments, increasing accountability. However, certain elements retain conditional clauses – for example, the requirement to respect alternative opt-outs depends on future standardisation or agreement. The definition of piracy websites also hinges on the ‘commercial scale’ qualifier and the publication of an official list.
Implementation timeline:
- From 2 August 2025 – New AI models deployed in the EU must comply with AI Act obligations. Signatories will benefit from a one-year grace period, provided they act in good faith towards compliance.
- From 2 August 2026 – Full enforcement begins, including potential fines.
- By 2 August 2027 – Models already on the market before August 2025 must comply.
The Code remains non-legally binding, and enforcement of copyright law remains the responsibility of national courts and the Court of Justice of the European Union.
At present, AI companies including Mistral and OpenAI have announced their intention to sign the Code. Stakeholders in the cultural sector continue to emphasise the importance of safeguarding authorisation, remuneration, and transparency across all AI-related policies, including the AI Strategy for Cultural and Creative Industries, the AI Continent Action Plan, AI Factories, and any potential revisions to the Copyright Directive.
- Get more insights in an article by Culture Action Europe
