EU AI labelling rules take effect this weekend with fines up to €15m

From Sunday, 2 August, any artificial intelligence system placed on the EU market must label synthetic images, audio, video and text that are designed to look authentic. The requirement stems from the transparency provisions of the EU AI Act, the bloc's landmark regulation that classifies AI by risk and imposes escalating obligations. New systems must comply immediately; those already in use have until December to adapt.

What the rules actually require

The obligation covers content that a reasonable person could mistake for genuine: a fabricated political speech, a manipulated photograph, a chatbot that presents itself as human. Each must carry a visible mark identifying it as AI-generated and, where technically feasible, an embedded digital watermark signalling its artificial origin. Text on matters of public interest, news articles, policy commentary, election material, must also be labelled if no human editorial oversight was involved.

Exemptions exist for content that is evidently artistic, satirical or fictional, and for personal user-generated material. Companies are encouraged, but not compelled, to label AI content created before the rules took effect. The European Commission has published black-and-white label templates that any provider may use, though custom designs are permitted provided they meet the visibility standard.

Penalties are significant: up to €15 million or 3% of worldwide annual turnover, whichever is higher. That places the transparency rules in the same enforcement tier as the Act's high-risk requirements for biometric identification and critical infrastructure management.

Industry warns of label fatigue

The Computer and Communications Industry Association (CCIA), whose members include Amazon, Google and Meta, argues that the Commission's July guidelines have stretched the definition of a deepfake far beyond what the 2024 Act envisaged. Boniface de Champris, CCIA Europe's AI policy lead, said the distinction between deceptive and benign synthetic content has been erased. "The label was meant to flag deceptive content. That distinction has been removed, so now almost everything gets labelled, a landscape in an advert is the same as a manipulated political speech."

He predicts a cookie-banner effect: once labels are ubiquitous, users stop registering them. "Like cookie banners, once labels are everywhere, users stop noticing them. That doesn't protect anyone, it just burdens commercial activity across the board." De Champris expects the most visible changes not on social platforms, which already label most AI material, but in advertising, film and publishing, sectors where AI use is widespread but invisible to audiences.

Big tech's existing defences

Several large platforms have built labelling infrastructure ahead of the deadline. TikTok says it requires creators to tag realistic AI-generated images, audio and video, and claims to have helped label more than three billion pieces of content. Independent research, however, has found AI-generated doctor personas accumulating millions of views on the platform while dispensing questionable health advice, suggesting enforcement gaps remain.

Google launched SynthID in 2023, an invisible watermarking tool that lets users verify whether an image or audio clip originated from its generative models. The company says it has applied the watermark to more than 100 billion images and 60,000 years of audio. Meta introduced policies in 2024 to identify photo-realistic AI images and says it has been developing industry-wide technical standards for AI audio and video detection.

A voluntary code to show good faith

Both Google and Meta are among more than 180 organisations that have signed a voluntary code of practice drawn up by independent experts under the supervision of the EU AI Office. Signing is treated by regulators as evidence of good intent; non-signatories must demonstrate compliance through other means. Markus Reinisch, Meta's vice-president for public policy in Europe, warned this week that a proliferation of competing labels could overwhelm users while adding regulatory complexity for providers. "We must avoid a situation where online content is subject to a growing array of different labels and disclosures that end up overwhelming people."

Political rationale: democracy, not just consumer protection

For the lawmakers who negotiated the Act, the stakes go beyond marketplace fairness. Sergey Lagodinsky, a Green MEP who helped steer the legislation, framed the rules as a democratic safeguard. He cited recent deepfakes, audio purporting to show Slovak opposition leader Michal Šimečka conspiring to rig an election, fabricated videos of French President Emmanuel Macron dancing in a 1980s nightclub, fake footage of then UK Labour leader Keir Starmer abusing staff, as evidence that synthetic media can distort public discourse before fact-checkers catch up.

Lagodinsky dismissed industry complaints as a familiar pattern. "I have a deep respect for industry, but in all too many cases, they complain, and then after things are implemented, it doesn't look that burdensome to them. And there is always a solution."

Enforcement begins in a fragmented landscape

The transparency rules are the first substantive obligations to take effect under the AI Act's phased rollout. National market surveillance authorities, coordinated by the Commission, will police compliance. Yet the practical mechanics remain unsettled: how watermarks survive compression and re-encoding, whether open-source models distributed outside commercial channels fall in scope, and how authorities will distinguish between a satirical meme and a deceptive political fabrication.

Henna Virkkunen, the Commission's executive vice-president for tech sovereignty, security and democracy, described AI as a transformative technology with extraordinary benefits but said the most advanced models "create risks on an entirely new scale". She argued the Act provides "a clear, risk-based and durable framework for trustworthy AI, one that gives innovators legal certainty while protecting the public interest."