AI-generated content: disclosure becomes mandatory on August 2

On 29 July the European Commission adopted its guidelines on transparency obligations for AI systems. The timing is no accident: on 2 August 2026 Article 50 of the AI Act becomes applicable, and with it the obligation to disclose chatbots, synthetic content and deepfakes. This is not just about big tech: it concerns anyone publishing AI-generated content — which, realistically, includes your company.

What Article 50 requires

Article 50 introduces four transparency obligations, split between those who provide AI systems (providers) and those who use them (deployers):

Declared interaction — people interacting with an AI system (a chatbot, a voice assistant) must know it, unless this is obvious to a reasonably well-informed and observant person.
Machine-readable marking — systems generating synthetic audio, images, video or text must mark their outputs in a machine-readable format, so they are detectable as artificially generated or manipulated.
Emotion recognition and biometric categorisation — people exposed to these systems must be informed, in compliance with the GDPR.
Deepfakes and text on matters of public interest — those publishing deepfakes, or AI-generated text informing the public on matters of public interest, must disclose it.

The information must be provided in a clear and distinguishable manner at the latest at the time of the first interaction or exposure, and in an accessible form.

Provider or deployer? SMEs are almost always deployers

The distinction matters, because the obligations differ. The provider develops and places the AI system on the market; the deployer uses it under its own authority. An SME running a chatbot on its website, generating articles and posts with generative AI tools, or producing synthetic promotional videos is — in the vast majority of cases — a deployer. And the labelling obligations for deepfakes and public-interest text, as well as the chatbot disclosure, fall precisely there.

In practical terms: the customer-service chatbot must present itself as AI; the promotional video featuring the CEO's synthetic avatar must be labelled; the industry article written by AI and published without human review must be disclosed.

The exemptions that really matter

Article 50 does not require labelling everything a generative model touches. The main exemptions:

Human editorial review — AI-generated text that undergoes human review, with a natural or legal person holding editorial responsibility, is exempt from the public labelling obligation. This is the most relevant exemption for those using AI as a writing tool.
Creative works — evidently artistic, satirical or fictional content carries lighter disclosure duties, to be fulfilled without spoiling the enjoyment of the work.
Assistive editing — tools assisting standard editing without substantially altering the content (correction, minor retouching) do not trigger the marking obligation.
Obviousness — if it is obvious you are interacting with an AI, no explicit notice is needed.

Beware of the flip side, though: the human-review exemption only works if the review actually happens and can be documented. "Someone probably read it" is not an editorial process.

The tools: guidelines and Code of Practice

The guidelines adopted on 29 July — the outcome of a public consultation with Member States, the AI Board and stakeholders — provide legal certainty on the scope of the obligations. Alongside them sits the Code of Practice on marking and labelling of AI-generated content, published on 10 June 2026: a voluntary instrument translating the obligations of Article 50(2), (4) and (5) into practical measures — digitally signed metadata, imperceptible watermarking, fingerprinting, logging — and visible labels, with a common EU icon set.

The logic mirrors the GPAI code of practice: signing up to the code is the facilitated route to demonstrating compliance; those who do not sign must prove to market surveillance authorities, on their own, that they meet the obligations through equivalent means. For a company buying technology, the question to ask vendors becomes: does your tool mark outputs in line with the Code of Practice?

The penalties

Non-compliance with the transparency obligations of Article 50 falls under Article 99(4) of the AI Act: administrative fines of up to €15 million or 3% of total worldwide annual turnover, whichever is higher. For SMEs and startups there is a proportionality corrective: the lower of the two amounts applies. Enforcement lies with national market surveillance authorities, alongside the AI Office and the European Data Protection Supervisor within their respective remits.

Five things to do now

Inventory outward-facing generative AI uses — chatbots, marketing content, images and video, press releases: where does AI touch the public?
Disclose your chatbots — a clear notice at first interaction, not buried in the terms of use.
Check your vendors — do the generation tools you use mark outputs in a machine-readable format? Have they signed the Code of Practice?
Define a labelling process — who decides whether a piece of content is a deepfake or public-interest text, and how it gets labelled before publication.
Formalise editorial review — if you want to rely on the text exemption, human review and editorial responsibility must be a documented process, not a habit.

Beyond the obligation

There is a short-sighted way to read Article 50: yet another compliance chore. And a far-sighted one: in an information ecosystem where synthetic content grows faster than our ability to recognise it, disclosing AI is an investment in credibility. Companies reaching 2 August with their processes in order are not just avoiding a fine: they are telling their customers that what they publish can be trusted.