Text, images and voices produced by artificial intelligence now reach screens in such volume that a simple question, who or what made this, has become one of the defining issues of the digital age. The European Union has chosen to answer it with law. Since 2 August 2026, the transparency obligations in Article 50 of the EU AI Act have applied across all 27 member states, turning disclosure of synthetic content from good practice into a legal duty.
The provision is one stage in the staggered rollout of Regulation (EU) 2024/1689, adopted in 2024. Article 50 covers four situations that touch the public directly: talking to an AI system, content an AI system generates, emotion recognition, and deepfakes.
Chatbots come first. Providers of systems built to interact with people, from customer-service bots to voice assistants, must design them so that users know they are dealing with a machine, unless that would be obvious to a reasonably observant person. Systems authorised by law to detect or investigate crime are exempt, with safeguards for the rights of others.
Generated content is the second pillar. Providers of systems that produce synthetic audio, images, video or text must mark the output in a machine-readable format so it can be detected as artificially generated or manipulated. Tools that only assist with standard editing, or that do not substantially change the input or its meaning, fall outside the rule. Systems already on the market before 2 August 2026 have a transition period until 2 December 2026 to add the marking.
The third set of duties falls on those who use the technology and publish its results. Anyone deploying a system to create deepfake images, audio or video must disclose that the material was artificially generated or manipulated. For evidently artistic, creative, satirical or fictional work, it is enough to flag the presence of such content in a way that does not spoil the work. AI-generated 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 holds editorial responsibility for it.
Emotion recognition and biometric categorisation systems carry their own rule: deployers must tell the people exposed to them how they work, and handle personal data under EU data protection law. In every case the information has to be clear and distinguishable, given no later than the first interaction or exposure, and accessible to people with disabilities.
To help companies comply, the Commission published a voluntary Code of Practice on the transparency of AI-generated content on 10 June 2026. Six independent experts named by the AI Office led the drafting, and its Annex I offers an optional EU icon, in three variants, for labelling AI content. On 20 July the Commission followed with non-binding guidelines on the scope, key terms and exemptions of Article 50, meant to steer the national authorities that enforce it.
Signing the code is optional and open to any provider or deployer active in, or planning to enter, the EU market; signatories file a form with the AI Office signed by someone able to bind the organisation. The code offers a streamlined way to show compliance, but it adds no obligations beyond the Act and does not replace it, and companies that stay outside must prove compliance by other appropriate means. Work on it began in September 2025 and went through three rounds of stakeholder consultation; the AI Office will review it at least every two years.
The reach goes well beyond Europe's own companies. The Act covers any provider or deployer that sells AI systems in the EU, or whose systems produce output used inside the bloc, although content published before the rules took effect does not need to be labelled retroactively. Companies that break the rules risk penalties reaching €15 million, or 3% of their global yearly turnover if that sum is larger.
