Companies that make or use AI will soon have to label content that looks authentic but is not, as the European Union’s landmark AI Act moves from principle into enforcement.

From 2 August, providers and deployers across the bloc face binding transparency rules covering deepfakes, synthetic media, and machine-generated text, a shift platforms have been bracing for since YouTube began auto-labelling AI videos.

The core idea is simple. If a piece of content could pass for real, people have a right to know a machine made it, and the burden now falls on the companies behind it.

The rules split the work in two. Providers, the firms that build generative systems, must embed machine-readable markings so that images, audio, video, and text can be detected as AI-generated or manipulated.

Deployers carry the visible half. Anyone who uses AI to create a deepfake, or to publish machine-written text on matters of public interest, must disclose that origin to the audience in a clear and timely way.

Intent does not get you off the hook. The duty to label a realistic deepfake applies whether or not the creator meant to deceive, closing a loophole that “we were only joking” might otherwise open.

Chatbots are covered too. When a person starts interacting with an AI system, they must be told at the moment of contact, not in the small print of a terms-of-service page nobody reads.

There are real carve-outs. Clearly unrealistic or fantastical content sits outside the rules, and AI-generated text escapes labelling if a human with genuine editorial responsibility has meaningfully reviewed it.

Creative work gets a lighter touch. Artistic and satirical content still has to carry a disclosure, but only in a way that does not spoil the experience of the work, a nod to film-makers and artists.

The penalties give the rules teeth. Breaches can draw fines of up to €15m or 3% of a company’s worldwide annual turnover, whichever is larger, enough to concentrate minds in Silicon Valley and Brussels alike.

The timing is not entirely fixed. The obligations formally apply from 2 August, but a separate simplification package has floated pushing the machine-marking deadline to December, part of a wider effort that thinned out parts of the AI Act.

Industry has been pushing at the edges. Advertisers, among others, have argued that AI-generated ads should be exempt from the transparency rules, a lobbying fight that has run alongside the law’s rollout.

The Commission has tried to add clarity. It published draft guidelines in May and is finalising a code of practice on how to mark and label AI content, which companies can sign to show they comply.

The reach is bigger than Europe. Because global platforms rarely build separate systems for a single market, the EU’s labelling regime is likely to shape how AI content is marked worldwide, much as its privacy law did before it.

Enforcement will be the hard part. Watermarks can be stripped, metadata can be lost, and machine-written text is notoriously difficult to detect, so the rules will only be as strong as the tools that check them.

Deepfakes are the sharpest test. The EU has separately moved to ban non-consensual intimate deepfakes, and the labelling regime is meant to catch the wider universe of synthetic media before it erodes trust entirely.

The stakes go beyond compliance. Europe is betting that mandatory transparency can preserve some shared sense of what is real online, at a moment when generative tools make convincing fakes trivial to produce.

Whether it works is an open question. The law is the most ambitious attempt yet to label the synthetic world, but its success depends on detection catching up with generation, a race the fakers have so far been winning.

Get the TNW newsletter

Get the most important tech news in your inbox each week.