Fake videos of politicians rigging elections, abusing their own staff, or dancing awkwardly in nightclubs have spread across the internet in recent years, engineered to smear their targets and mislead voters. The European Union is now moving to slow that flood.

From Sunday, artificially generated images, audio, and text designed to pass as authentic must carry a label across the EU. The requirement falls under the bloc's AI Act, and it forces companies to make sure people know when they are talking to an AI chatbot or looking at a manipulated image or a machine-written text.

How the Rules Work and Who They Cover

New AI systems entering the EU market from August 2 fall under the rules immediately. Systems already on the market get an extra four months to comply.

Synthetic text, images, video, and audio built to look truthful must be visibly marked as AI-generated and carry a digital watermark that signals its artificial origin. Texts on matters of public interest also have to be labelled as AI-made when no human editor has overseen them.

There are limits to the reach. The rules do not touch users' personal content, and there is a carve-out for works that are evidently artistic, satirical, or fictional. Companies are encouraged to label AI content that predates the rules, though doing so is not required.

To make compliance simpler, the European Commission has released black-and-white labels anyone can use, and companies remain free to design their own.

The Penalties for Getting It Wrong

Non-compliance carries real financial weight. A company that fails to meet the requirements faces fines of up to 15 million euros, or 3% of its worldwide annual turnover, whichever regulators decide to apply.

A Democracy Argument, Not Just a Consumer One

For the lawmakers who wrote the AI Act, the stakes go beyond shopping habits. Green MEP Sergey Lagodinsky, who helped negotiate the legislation, framed the transparency requirement as a safeguard for public life itself.

“It is a matter not only of customer protection, it's also a matter of democracy protection,” Lagodinsky told the Guardian. “Making transparent this information is something which we need to preserve our democracy and the authenticity of facts online.”

The examples he and others point to are concrete. Fabricated audio purported to catch Slovakian opposition leader Michal Šimečka scheming with a journalist to rig an election. Mocked-up footage placed French president Emmanuel Macron dancing in a 1980s nightclub. Another fake showed then-UK Labour leader Keir Starmer abusing party staff.

Industry Backs the Goal but Fears Label Fatigue

The tech sector says it supports making deepfakes visible. What it worries about is scope. Companies fear the rules will be read so broadly that ordinary internet users end up buried under banners and warnings.

The Computer and Communications Industry Association argues that EU guidelines published in July stretched the definition of a deepfake well past the boundary set in the 2024 AI Act.

“The label was meant to flag deceptive content,” said CCIA Europe's AI policy lead, Boniface de Champris. “That distinction has been removed, so now almost everything gets labelled – a landscape in an advert is the same as a manipulated political speech. 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 biggest shift to land away from social platforms, which already flag most AI content. The change people will actually notice, he said, will show up in advertising, film, and publishing, sectors where AI is used heavily but rarely disclosed. The heaviest burden, in his view, falls on those least familiar with AI regulation.

What the Platforms Already Do

Several of the largest tech firms already run their own disclosure policies. TikTok says it requires creators to label realistic AI-generated images, audio, and video, and reports having helped label more than 3 billion pieces of content. Even so, recent research found AI-generated “doctors” racking up millions of views on the platform while pushing dubious health advice.

Google launched a tool called SynthID in 2023 that lets users check whether an image is AI-generated. The company says it has embedded an invisible watermark on more than 100 billion images and on 60,000 years' worth of audio.

Google and Meta have both signed a voluntary code of practice meant to help companies meet the EU's transparency rules. The Commission said more than 180 organisations had signed the code, which independent experts drew up under the supervision of the EU AI office. Regulators treat signing as a signal of good intent, while those who do not sign must prove they are meeting the rules some other way.

Meta, which owns Facebook and Instagram, rolled out policies in 2024 to help people identify photo-realistic AI images and says it has been developing industry-wide standards for AI audio and video. Announcing the company's intention to sign the EU code on July 28, its vice-president for public policy in Europe, Markus Reinisch, warned against overload: “We must avoid a situation where online content is subject to a growing array of different labels and disclosures that end up overwhelming people, while at the same time adding additional regulatory complexity for providers.”

A Wider Enforcement Push Begins

The transparency measures are one piece of broader AI Act enforcement kicking in from August 2. The Commission's lead official on tech policy, Henna Virkkunen, described AI as a transformative technology capable of extraordinary benefits, while cautioning that the most advanced models create risks on an entirely new scale.

“Europe anticipated this development. With the AI Act, we established a clear, risk-based and durable framework for trustworthy AI - one that gives innovators legal certainty while protecting the public interest,” Virkkunen said.

Lagodinsky, for his part, expects the industry to adapt despite its objections. “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.”