The Code of Practice on Transparency of AI-generated Content: sign or not?

The day before Article 50 became applicable, the European Commission announced “strong backing” for the Code of Practice on Transparency of AI-generated Content — a voluntary code that maps the marking and labelling obligations of Article 50(2), (4) and (5) of Regulation (EU) 2024/1689 onto concrete measures. About 190 organisations had signed by the end of July 2026, per the Commission, and procurement teams are starting to ask vendors whether they are on the list. Here is what the code actually is, what signing does (and does not do) for your Article 50 compliance, and how to decide.

What the code is

The code was drawn up by independent expertsin a multi-stakeholder process facilitated by the EU AI Office, and has been assessed as adequate by both the European Commission and the AI Board. It is published on the Commission's official page (full text, PDF) and has two sections:

• Section 1 — Providers. Rules for marking and detection of AI-generated and manipulated content: the machine-readable marking duty of Article 50(2).
• Section 2 — Deployers. Rules for labelling of deep fakes and AI-generated or manipulated text published on matters of public interest: the visible-disclosure duties of Article 50(4).

Organisations sign only the section(s) that apply to them. Although the Article 50(2) legal obligation binds providers, Section 1 can also be signed by providers of marking and detection solutions and providers of AI models that help generative-AI providers implement the mandatory measures (Commission announcement).

What signing actually gets you

Signing does not create new obligations and not signing does not remove any: Article 50 applies from 2 August 2026 either way. What the code changes is how you demonstratecompliance. In the Commission's own words, signatories “can rely on its measures to demonstrate compliance” and get a streamlined, predictable and legally certain pathway to guarantee and demonstrate compliance — regardless of where the organisation operates or which supervisory authority is competent.

In practice this matters in three situations:

1. Regulator contact.If a market surveillance authority asks how you meet Article 50(2) or 50(4), implementing the code's measures is a recognised answer instead of a bespoke legal argument.
2. Procurement questionnaires. Buyers increasingly ask suppliers of generative-AI features whether they have signed. A signature is an easy, checkable signal; the substance behind it is whether you actually implement the measures.
3. Cross-border consistency. Enforcement is national (see our country-by-country enforcement table), but the code is one common yardstick across all Member States.

What the signatory list tells you

The Commission publishes the full list of signatories and updates it on an ongoing basis. At the time of the Commission's announcement (late July 2026), the page listed 83 Section 1 signatories and 152 Section 2 signatories. The list ranges from major model providers (e.g. Anthropic, Cohere, Aleph Alpha, Black Forest Labs) to SMEs, media companies and public-sector bodies.

Three takeaways from the list:

• It is not just Big Tech. The majority of signatories are European SMEs and organisations that deploy — rather than build — generative AI. Section 2 has more signatories than Section 1, which reflects how many more companies are deployers than providers.
• Buyers will use it as a filter.Because the list is public and searchable, expect “are you a signatory?” to appear in vendor due-diligence checklists the same way SOC 2 or ISO 27001 questions do.
• Absence is not non-compliance. The code is voluntary; a company can be fully Article 50-compliant without signing. But if you are not on the list, be ready to show equivalent evidence of how you mark and label AI content.

Sign or not? A decision rule

• You provide a generative AI system (including white-labelled or API-wrapped ones) whose output reaches EU users: signing Section 1 is low-cost and pre-empts the compliance question — if you can implement the marking and detection measures. Do not sign ahead of your actual capability.
• You deploy generative AI and publish deep fakes (e.g. synthetic avatars, voice clones) or AI-written public-interest text: Section 2 formalises labelling you already owe under Article 50(4).
• You only use AI internally with no EU-facing synthetic output: there is nothing to sign; Article 50(2)/(4) duties are unlikely to apply to you in the first place.

The AI Office keeps signing open — the procedure and forms are on the Commission's “How to sign” page. Alongside the code, the Commission also published a set of common labelling icons that deployers may use to label AI-generated content.

Whether or not you sign: the obligations stand

The code is a compliance pathway, not the obligation itself. Since 2 August 2026, chatbot disclosure, machine-readable marking and deep-fake labelling are enforceable law, with fines up to EUR 15 million or 3% of worldwide turnover (Article 99(4)(g)). If you have not yet mapped which paragraphs of Article 50 apply to you, run the free 3-minute check — and if you need marking, badges or evidence packs to implement the measures, see our plans. Note the transitional window for systems already on the market before 2 August 2026: the Article 50(2) marking grace period runs to 2 December 2026.