MOZOM vergelijkt

EU labels for AI: protection for citizens or a new switch on the flow of information?

Europese Commissie, AI Act Service Desk, DSA Transparency, OpenAI, The Guardian, Financial Times en Reclaim The Net Technology 1 augustus 2026 om 08:55

Type: Source comparison Author: Paul Giezen Published: 1 augustus 2026 om 08:55 Report a correction

Realistic editorial image of a European digital control room with abstract AI content labels and a platform dashboard, illustrating AI Act and DSA oversight.

Source: Europese Commissie, AI Act Service Desk, DSA Transparency, OpenAI, The Guardian, Financial Times en Reclaim The Net
MOZOM headline: EU labels for AI: protection for citizens or a new switch on the flow of information?
Original headline: From 2 August 2026, EU transparency obligations apply to certain AI systems while DSA oversight is already formalising the platform layer
Author: Paul Giezen
Date: 1 augustus 2026 om 08:55
Subject: MOZOM examines how the new EU transparency obligations for AI from 2 August 2026 intersect with DSA oversight of major platforms, and why labels can become both protection and a tool of power.

Summary of the original report

The European Commission published guidelines on 20 July 2026 for the transparency obligations in Article 50 of the AI Act. Those obligations apply from 2 August 2026. AI providers must inform users when they interact directly with AI. Providers of AI systems that generate synthetic audio, images, video or text must also in principle mark output technically and in machine-readable form as artificially generated or manipulated. For publishers and other professional users, the obligation mainly concerns clear disclosure when they publish deepfakes or AI-generated text on public-interest matters without human editorial control. Deployers of certain AI systems must inform people about emotion recognition, biometric categorisation, deepfakes and AI text on public-interest matters without human editorial control. Officially, the goal is transparency, trust and protection against deception. The critical layer is that these rules do not stand apart from the Digital Services Act. Large platforms must already analyse risks, publish transparency reports, undergo audits and maintain notice and appeal procedures. Together, the AI Act and DSA therefore form not only consumer protection, but also a more formal infrastructure for classification, visibility and enforcement of online information.

How do different sources present it?

The European Commission presents Article 50 mainly as practical transparency: citizens should know when they are dealing with AI and when realistic-looking content is artificial. The Guardian stresses mandatory labels and fines of up to 15 million euros or 3 percent of global annual turnover. Financial Times describes it as a possible cookie-banner moment for AI: visible to everyone, but late and sometimes vague for companies. Reclaim The Net takes the harder civil-liberties reading and sees DSA expansion around major AI and gaming platforms as a risk for regulated online speech. The sources do not disagree about the existence of rules, but they do differ over their meaning.

Where does the emphasis differ?

There are three reasonable readings. The protection reading says citizens must be able to see whether they are dealing with AI, deepfakes or synthetic public information. The compliance reading says companies receive a new administrative layer of labels, technical markings, reports and fine risk. The power reading says that once governments, regulators and platforms jointly determine which content must be labelled, restricted, removed, reported or investigated, an administrative switch emerges on the flow of information. That third reading does not prove total censorship, but it is not fantasy either. The official documents themselves describe reporting mechanisms, risk assessments, audits, platform reports and enforcement.

Own source research: four layers that together form a control chain

MOZOM placed the AI Act documents beside the DSA documents. Layer one is recognisability: users must know when they interact directly with AI. Layer two is detectability: providers of generative AI systems must machine-readably mark synthetic output as far as technically feasible. Layer three is public disclosure: professional users must clearly disclose deepfakes and AI text on public-interest matters without human editorial control as artificial. Layer four is platform oversight: under the DSA, very large platforms and search engines must analyse risks, report, undergo audits and maintain systems for notices, appeals and government contact points. The added value is in the combination. Separately, each part sounds reasonable. Together, they show that online information is filtered not only by the creator, but also by detection systems, compliance departments, national authorities, the European Commission and platform rules.

Protection sounds neutral, classification is power

The word transparency has a clean image. Nobody wants to be deceived by a deepfake or by a chatbot pretending to be human. Still, the political question is broader. A label is never only a sticker. It is a judgement attached to information. If that judgement later affects recommendations, reach, ad access, search results or moderation, a transparency rule can in practice become a visibility rule. That is not written everywhere in the AI Act, but it sits in the overlap with platform oversight.

Why this is sensitive now

The EU is building several digital legal layers at the same time: the AI Act, the DSA, the DMA and additional transparency obligations. In the same period, news consumption is moving from websites to platforms, search engines and AI answers. That makes the question of who controls the gatekeepers more important than ever. OpenAI itself reports 159.1 million average monthly active recipients for ChatGPT Search in the EU over the six months ending 31 March 2026. That figure was published for DSA purposes and should not be used for everything, but it does show how large the AI information layer already is. If such systems start mixing news, search results, summaries and labels, transparency becomes both necessary and powerful.

Possible message behind the news

The official message is: AI must be recognisable, platforms must become safer and citizens receive more rights. The critical counter-message is: the same infrastructure can also determine which information appears visible, reliable or suspicious. The line between protection and steering does not run through a press release, but through the everyday operation of labels, search engines, recommendations and moderation.

Neutral conclusion

The conclusion is not that every AI label is censorship. That would be too easy. The tougher conclusion is that Europe is building a new digital governance layer above the flow of information. It can protect citizens against deepfakes and manipulation, but it can also concentrate power among regulators, platforms and technical detection systems. Anyone who takes freedom of information seriously should therefore not only ask whether AI is labelled. The real question is: who labels the labeller?