A signature just before the rules arrive

Google has chosen the European Union’s shared route for showing when content was made or changed by AI.

The company said on Friday that it is signing the Code of Practice on Transparency of AI-Generated Content. The code gives providers and professional users of generative AI a practical set of measures for meeting the transparency duties in Article 50 of the AI Act.

The timing is tight. Organisations have until 27 July at 18:00 CEST to appear on the initial list of signatories. The legal obligations begin to apply on 2 August.

One distinction matters here. Signing the code is voluntary. Complying with Article 50 is not. A company can choose another route, but it will have to show regulators that its own measures are adequate.

Google’s announcement is therefore both a policy commitment and a compliance choice. It gives the company a recognised framework across the EU instead of asking each national authority to judge a different approach from scratch.

What has to be marked, and what has to be labelled

The rules split the work between the company that provides an AI system and the organisation that uses it.

Providers of systems that generate synthetic audio, images, video or text must make the output machine-readable and detectable as artificially generated or manipulated. The method should be effective, interoperable, robust and reliable as far as technically feasible. Editing tools that do not substantially change the original material can fall outside that duty.

Deployers have a more visible job. Deepfakes must be disclosed. AI-generated or manipulated text published to inform the public on matters of public interest must also be disclosed, unless it has gone through human review or editorial control and a person or organisation takes editorial responsibility.

The code translates those broad legal requirements into two sections: one for providers and one for deployers. The EU has also made a set of icons available for labelling, although using those icons is not the whole of compliance.

This is less like putting one sticker on every AI output and more like building a chain. A technical mark has to survive the creation process, platforms need to carry useful provenance, and the person seeing the content still needs a clear notice.

Google already has some of the plumbing

Google points to SynthID, its system for embedding and detecting signals in AI-generated media, and to the C2PA provenance standard. It says signing the code is consistent with both efforts.

The company also says it has been working with Apple, ElevenLabs, Kakao, Nvidia and OpenAI on interoperable watermarking tools that use SynthID. Those are Google’s descriptions of its partnerships and technology; Model Current has not independently tested how well the marks survive editing, compression or movement between platforms.

That gap is important. A watermark that works inside one product is helpful. A signal that stays useful after a file has been copied, cropped, re-encoded and posted elsewhere is much harder.

The EU code recognises that technical methods are still developing. It asks for effective and interoperable solutions, but ties that duty to technical feasibility, cost and the current state of the art.

Signing gives Google a common target. It does not prove that every Gemini image, edited clip or synthetic voice will arrive with a durable explanation in every place a person encounters it.

Google supports the code and questions the label pile-up

Google’s note is not a simple endorsement.

The company says added regulatory complexity could work against Europe’s goal of simplification while technical solutions are still changing. Its sharper concern is about the user experience: too many overlapping AI labels and legal disclosures may make the context harder to understand.

That is a reasonable problem to test, not a reason to skip disclosure. People already see cookie notices, sponsored labels, edited-media warnings and platform-specific symbols. Adding another mark only helps if its meaning is consistent and the important information is easy to find.

The European code is meant to reduce that fragmentation by creating common practices. Google’s warning points to the next stage. Regulators and signatories will need to learn whether the notices are noticed, understood and trusted, rather than counting how many were displayed.

Clear provenance can support trust. A crowded interface full of badges can become background noise. Both outcomes are possible.

What is confirmed, what Google says and what remains open

The confirmed facts are narrow. Google has announced that it will sign the code. The code is voluntary, the related Article 50 duties are legal, the initial-signatory deadline is 27 July and the duties apply from 2 August.

Google says SynthID, C2PA adoption and its work with other technology companies put it on a practical path to implementation. It also says overlapping notices could confuse people. These are company positions, not findings from an independent audit.

Several details are still open. The EU has not yet published the initial signatory list. Google has not set out a product-by-product timetable in this announcement. There is no public measure of how well its proposed labels will survive common edits or how accurately people will interpret them.

There is also a proposed transition until 2 December for some systems already on the market, but the Commission’s own guidance describes that as part of an AI Omnibus proposal, not a finished rule.

The signature is useful because it moves the discussion from whether to disclose towards how. The real test starts when the marks leave the lab and meet an ordinary feed.

Sources

  1. Google — Signing the EU AI Act Transparency CodePrimary company announcement published 24 July 2026. Source for Google’s decision, its SynthID and C2PA position, named collaborations and concern about overlapping labels.
  2. European Commission — Code of Practice on Transparency of AI-Generated ContentPrimary policy page explaining the code’s two sections, its voluntary status, practical effect and relationship with the AI Act.
  3. European Commission — Signing the CodePrimary FAQ for the 27 July initial-signatory deadline, legal effect, alternative compliance routes and the proposed transition for some existing systems.
  4. European Commission — Article 50 transparency guidelinesPrimary implementation guidance adopted 20 July 2026, confirming scope and the 2 August application date.
  5. EU AI Act Service Desk — Article 50Official article text for provider marking, deployer disclosure, exceptions, accessibility and the role of codes of practice.