EU makes AI labels mandatory on authentic-looking images, audio and text

Starting August 2, the European Union will require companies to disclose when images, audio and text have been artificially generated or manipulated to look authentic, as the Guardian reports. The rules are part of the EU’s AI Act, and they could have a particular impact on publishers and other organizations that are increasingly using AI to produce content.

Under the new requirements, people must be told when they are interacting with an AI chatbot or viewing manipulated media. Synthetic content must carry a visible label as well as a digital watermark identifying it as AI-generated.

The rules apply to new AI systems entering the European market beginning August 2. Existing systems will have an additional four months to comply. And for publishers, there is another important provision: Text about matters of public interest must be labeled as AI-generated when there has been no human editorial oversight.

Companies that fail to comply could face penalties of up to €15 million.

Are you an AI Agent? Read more here >

There are some exceptions. Personal content is exempt, as are works that are clearly artistic, satirical or fictional. Labeling AI-generated content created before the rules took effect is encouraged, but not required.

The European Union says the goal is to make it easier for people to distinguish authentic material from synthetic content. The rules follow the June 10 publication of a voluntary Code of Practice on Transparency of AI-Generated Content, which lays out how AI providers and deployers can comply with the AI Act’s labeling requirements.

Sergey Lagodinsky, a Green member of the European Parliament, told the Guardian that the rules are about more than protecting the public.

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

But some technology companies and industry groups say the rules have become too broad.

The Computer and Communications Industry Association argues that guidelines issued in July expanded the definition of a deepfake beyond what lawmakers intended when they passed the AI Act in 2024.

“The label was meant to flag deceptive content,” said Boniface de Champris, CCIA Europe’s AI policy lead. “That distinction has been removed, so now almost everything gets labeled — a landscape in an advert is the same as a manipulated political speech… That doesn’t protect anyone, it just burdens commercial activity across the board.”

De Champris says the biggest changes may not be obvious on social media, where platforms already label much AI-generated material. He expects the rules to have more consequences for advertising, film and publishing — industries that are using AI at scale but do not always tell audiences when they do.

That could put news organizations and other content producers in the EU in a new position. AI-generated images, synthetic voiceovers and machine-written text about matters of public interest may now require disclosure under the rules. What was once a question of editorial policy could become a matter of compliance.

Some major technology companies have already taken steps in that direction. TikTok says it requires creators to label realistic AI-generated media and has labeled more than 3 billion pieces of content. Google says its SynthID watermarking system has been used on more than 100 billion images. And more than 180 organizations, including Google and Meta, have signed a voluntary code of practice supporting the EU’s transparency requirements.

Henna Virkkunen, the European Commission’s executive vice president for tech policy, says the broader rules are meant to strike a balance. She described AI as a “transformative technology” capable of bringing “extraordinary benefits,” while warning that the most advanced systems could create risks “on an entirely new scale.”

For publishers and other content producers, the next phase will be figuring out exactly where those lines fall — and what AI disclosure looks like when it becomes a legal requirement rather than an editorial choice.