What the labelling duty and the start of enforcement cover
The AI Act's transparency obligations are the rules that make users of AI-powered services able to tell that what they are seeing came from AI. On August 2, 2026, those rules became applicable and enforcement of the Act itself began at the same moment. This was the day an obligation written in the text became an obligation authorities can act on.
The three labelling rules that apply from August 2, 2026
| Rule | Who it covers | What it requires |
|---|---|---|
| Say it is AI | Chatbots and other interactive AI | Tell users they are dealing with AI, not a human |
| Label it | Deepfakes (images, video, audio generated or edited with AI) | Attach a label making clear the content was created |
| Machine-readable mark | AI-generated or altered content generally | Embed a mark that can be detected |
The three rules that started on August 2
The Commission lists three. First, chatbots and other interactive AI systems have to tell users they are dealing with AI, not a human. Second, deepfakes — images, videos, or audio edited or generated using AI — have to be labelled. Third, AI-generated or altered content has to carry machine-readable marks so it can be detected more easily.
The stated aim is to reduce deception and manipulation and help people make informed choices. From the business side, obligations that are clearer also come with a clearer way of showing compliance. For the full application schedule and the high-risk delay, see what actually starts on August 2 under the EU AI Act.
From 2 August 2026, the European Commission’s AI Office, together with national authorities, will begin enforcing the Artificial Intelligence (AI) Act. / Under the new rules, chatbots and other interactive AI systems will have to tell users they are dealing with AI, not a human. Deepfakes (images, videos, or audio that have been edited or generated using AI) will have to be labelled. AI-generated or altered content will also have to carry machine-readable marks so it can be detected more easily. / The measures are intended to reduce deception and manipulation and help people make informed choices. They also give businesses clearer obligations and a practical way to show compliance. — From the start of enforcement, the three labelling rules, and the stated aim of the measures
The signatory list was published on the same day
Alongside the start of enforcement, the Commission published for the first time a list of organisations that signed the Code of Practice on Transparency of AI-generated Content. The first list runs to more than 180 organisations, while the code's policy page states that about 190 had signed by the end of July 2026. The two figures differ, and the reason for the gap has not been published.
The code was drawn up by independent experts in a multi-stakeholder process facilitated by the AI Office. The point is to turn the labelling rules into shared working procedure instead of leaving every company to read the provisions on its own.
The Code of Practice on Transparency of AI-generated Content was drawn up by independent experts in a multi-stakeholder process facilitated by the AI Office. / By the end of July 2026, about 190 companies organisations have signed the code. — From the drafting process and the number of signatory organisations (the phrase `about 190 companies organisations` appears as written on the source page)
What changes when you sign the transparency code
The code sets out working procedure matching the obligations in Article 50(2), (4) and (5). Signing does not lighten the obligation. What it changes is only how you show you are meeting it.
The code's two sections, and what comes with signing
One section for providers, one for deployers
The code splits in two. Section 1 addresses providers — the side that builds — and covers marking and detection of AI-generated and manipulated content. Section 2 addresses deployers — the side that uses — and covers labelling of deepfakes and AI-generated or manipulated text. The side that builds embeds the mark; the side that uses displays the label. That division is the chapter structure.
The EU has also produced a set of icons that deployers of generative AI systems may use to label AI-generated content. Signatories join Signatory Taskforces where implementation practice is shared.
The Code of Practice on transparency of AI-generated content (PDF) has 2 sections : Section 1: Providers - Rules for marking and detection of AI-generated and manipulated content Section 2: Deployers - Rules for labelling of deepfakes and AI-generated and manipulated text / The EU has also created a set of icons that deployers of generative AI systems may use to label their AI-generated content. / Signatories to the code will also collaborate in the Signatory Taskforces, which will be set up to share practices and advance the implementation of the marking and labelling. — From the two-section structure, the official labelling icons, and the Signatory Taskforces
Signing settles how compliance is demonstrated
The Commission and the AI Board have confirmed the code is an adequate voluntary tool for demonstrating compliance with the transparency obligations. Following that positive assessment, providers and deployers who sign can rely on the code's measures to demonstrate compliance. The Commission's stated payoff is reduced administrative burden plus predictability, legal certainty and trust across all Member States.
What signing does not do is remove the Article 50 requirements. Adherence stays voluntary, and the transparency requirements remain legal obligations.
It helps providers and deployers of generative AI systems to comply with the AI Act’s obligations for labelling and marking of AI-generated content – Article 50(2), (4) and (5) of the AI Act. Even though adherence to the code is voluntary, the transparency requirements under article 50 of the AI Act are legal obligations. The Commission and the AI Board have confirmed that the code is an adequate voluntary tool to demonstrate compliance with the AI Act transparency obligations. / Following a positive assessment of the code, all providers and deployers who sign it can rely on its measures to demonstrate compliance with the AI Act’s rules for labeling and detection of AI-generated content, deepfakes and certain text publications. This will reduce their administrative burden and give them predictability, legal certainty and trust across all Member States. — From the provisions covered, the relationship between voluntary adherence and legal obligation, and the effect of signing
Not signing means proving adequacy yourself
Complying through other means is allowed. But then you have to demonstrate that the measures you chose are adequate. That is assessed individually by market surveillance authorities in each Member State, which leaves room for the same implementation to be judged differently across borders. Use a common template that travels across the EU, or take on country-by-country assessment. In practice those are the two options.
By contrast, providers and deployers that decide to comply through other means will have to demonstrate that those measures are adequate. This will be assessed individually by different market surveillance authorities. — From the demonstration burden when complying by other means and the individual assessment by market surveillance authorities
The code itself is published as a PDF and it runs long. If you only need specific provisions, converting it to Markdown first keeps the headings and lists intact while you pull them.
Where complaints go, and what to check if you are outside the EU
Together with the start of enforcement, the Commission set out where breaches can be reported. What matters in practice is less the enforcement apparatus than the fact that inbound reporting routes opened at the same time.
Complaint and whistleblowing channels the Commission points to
| Channel | Who it is for |
|---|---|
| AI Act complaints tool | Users and organisations who spot a breach |
| AI Act Whistleblower Tool | People reporting from inside a business |
| Complaints channel for downstream providers | Businesses building on general-purpose AI models |
Three channels opened at once
In its enforcement material the Commission lists the AI Act complaints tool, the AI Act Whistleblower Tool, and a complaints channel for downstream providers using general-purpose AI models. Once reporting routes exist officially, breaches get brought to authorities rather than waiting to be found. The same material points to the guidelines on transparency of AI-generated content and the guidelines for providers of GPAI models.
Read more about the Enforcement of the AI Act and the: AI Act complaints tool AI Act Whistleblower Tool Complaints channel for downstream providers using general-purpose AI models / Find more information about: Guidelines on Transparency of AI-generated Content General-Purpose AI (GPAI) Code of Practice Guidelines for providers of GPAI models Guidelines on prohibited AI practices — From the list of channels and guidelines given as related information for enforcement
Two things to check first from outside the EU
For a business outside the EU serving EU users, two checks are worth more than the rest at this stage. One is whether the generative AI you build on has signed the code. If it has, marking gets implemented on the provider's side against a shared template. The other is labelling, if you are the one publishing AI-generated content — the deployer role. Given that official EU icons exist, checking whether you can use those beats designing your own notation.
For how the framework got here, including the delay to high-risk obligations, see the AI Omnibus explainer.
Conclusion: check your provider's signature and your own labelling
On August 2, 2026 the EU AI Act became an enforced regulation and the labelling rules for AI-generated content started running. The practical footing is a code of practice with around 190 signatories, structured as marking for providers and labelling for deployers. Signing is voluntary. Not signing means demonstrating to national authorities that your alternative measures are adequate. If you serve the EU, start by confirming whether your provider has signed. If you publish AI-generated content yourself, start with the official EU icons. That is enough for now. When you read the code PDF, converting it to Markdown first keeps you from losing the article numbers as you pull the relevant parts.
