Skip to content

LeadershipNot yet confirmed elsewhere1 publisher2 min readPublished

The EU's AI transparency code is voluntary. The 2 August 2026 labelling duty is not.

The code is optional. The marking and labelling duty it implements is not, and the alternative to signing is proving your own method to every national regulator that asks.

The Board Room · Leadership desk

How we use AISend a correction

What happened

  • Article 50's marking and labelling rules for providers and deployers of generative AI systems apply across the EU from 2 August 2026.
  • The Commission and the AI Board have confirmed the transparency code of practice as an adequate voluntary tool for demonstrating compliance.
  • About 190 companies and organisations had signed by the end of July 2026, according to the Commission.
  • The code splits in two: provider rules on marking and detection, deployer rules on labelling deepfakes and AI-generated or manipulated text.
  • Adherence is voluntary; the Article 50 requirements the code implements are legal obligations.

Why it matters

  • cost Staying off the list removes no engineering work and adds legal work: each market surveillance authority judges an alternative method on its own, and the compliance function pays for that repetition.
  • exposure The duty reaches whoever puts the content out, so a team labelling deepfakes or AI-generated text is in scope even when the model belongs to someone else.
  • decision With the count taken days before the obligation attached, any firm not on the list on day one has already chosen the self-justification route rather than deferred the choice.
  • constraint Because the EU's icons are offered rather than required, each deployer can design its own mark, and readers get no single signal to learn to look for.

A signature does not lighten the duty; it settles who gets to argue about it. The measures in the code carry recognition across every Member State, and that portability is what is actually on offer: predictability, legal certainty and a lighter administrative load in all of them at once [5]. Nothing about the product changes.

The recognition has edges. The code operationalises Article 50(2), (4) and (5) [7], which is three paragraphs of an article that covers more ground than that, so signature evidences compliance for those parts and not the rest [16]. The code also does not replace the AI Act or the Commission's guidelines on the scope of Article 50 [12], and those guidelines sit alongside it as the definition of what is in scope [14]. Treating a signature as the whole compliance answer misreads what was signed.

Nor is signing a one-time filing. Signatories join Signatory Taskforces, set up to share practices and advance implementation of the marking and labelling rules [11]. That is standing staff time, and it is also the venue where the working meaning of adequate marking gets settled. Consider who drafted the text in the first place: providers of generative AI systems, developers of marking and detection techniques, associations of deployers, civil society organisations, academic experts and organisations with expertise in transparency and very large online platforms, in a process facilitated by the AI Office [13]. The firms selling provenance and watermarking tooling helped describe what compliant marking looks like. That is ordinary for technical codes, and it is also a reason to read the detection measures as a market position rather than a neutral floor.

For anyone shipping generative output into Europe and not on the list, the exposure is procedural rather than technical. The marking has to exist either way; what the non-signatory lacks is a document a regulator has already agreed to accept [5]. That gap is cheap to close while the questions are hypothetical and expensive to close once a market surveillance authority has opened a file on a specific product [6]. And because the Taskforces are where implementation practice will be worked out among signatories [11], the companies outside them will receive the definition of adequate marking rather than help write it. The code is voluntary in the same sense that filing your tax return on the official form is voluntary.

What to watch

  • Whether the signatory list moves materially past 190 once the obligation is live, and which large providers stay off it.
  • The first market surveillance assessment of a non-signatory's alternative measures, and whether two Member States reach the same verdict on comparable methods.
  • Whether the EU icon set is adopted widely enough to become the expected mark, or fragments into house styles.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence72
Adoption52
Hype gap+8
Incentives68
Confidence62
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    The AI Act's Article 50 transparency obligations for providers and deployers of generative AI systems are applicable from 2 August 2026.

    ReportedSupportedSource: European Commission, Shaping Europe's digital futureView cited source
  2. [2]

    The Article 50 obligations pertain to marking and detection of AI-generated content and labelling of deepfakes and certain AI-generated publications.

    ReportedSupportedSource: European CommissionView cited source
  3. [3]

    Adherence to the Code of Practice on Transparency of AI-generated Content is voluntary, but the transparency requirements under Article 50 of the AI Act are legal obligations.

    ReportedSupportedSource: European CommissionView cited source

Sources

1 independent publisher whose own reporting we read for this story.

  1. digital-strategy.ec.europa.eu

    1 article · August 22, 2026

    Code of Practice on Transparency of AI-generated Content | Shaping Europe’s digital future

Share your take

Let Clarity write the post for you.

Signed-in readers get a short post drafted on this story in the register they choose — narrative, analytical, or a direct position — editable to the last word before it goes anywhere. The share buttons at the top of this story work without an account.

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

  • Voluntary codes as compliance routesFollow
  • EU AI Act Article 50 transparency dutiesFollow
  • AI content provenance and watermarkingFollow
  • Deepfake and AI text labellingFollow

Entities

Loading related stories