EU Deepfake Labeling Rules: What They Mean For Your Reputation

Since 2 August 2026, anyone who publishes a deepfake in the European Union has had a legal duty to say so. Article 50 of the EU AI Act now requires AI-generated and AI-manipulated content to be labeled, chatbots to identify themselves, and generative AI providers to mark their output in a machine-readable way. A second deadline arrives on 2 December 2026, when the grace period for older generative systems runs out.

For compliance teams, this is a labeling project. For anyone responsible for a brand or a public profile, it is something bigger: a new set of rules that changes how fake content about you is made, spotted and removed, and how your own use of AI will be judged.

What Article 50 Actually Requires

Article 50 sets four transparency duties, and they apply to any AI system used in these situations, not only to systems classed as high risk:

• Chatbots must say they are AI. People have to be told when they are talking to a machine.
• Providers must mark synthetic output. Companies that build generative tools have to embed machine-readable markers in the images, audio, video and text their systems produce.
• Deployers must label deepfakes. Whoever publishes AI content that resembles real people, places or events, and would falsely appear authentic, must disclose that it was artificially generated or manipulated, at the latest when someone first sees it.
• AI-written public interest text must be disclosed. Text on matters of public interest needs a label unless a human has genuinely reviewed it and an identifiable person or organization takes editorial responsibility.

The European Commission published its final guidelines on 20 July 2026, and they read the deepfake definition broadly. It now covers digital replicas and distinctive personal characteristics, and subjects that exist or could plausibly have existed. The guidelines also make clear that the label is required whether or not anyone intended to deceive, and that a spell-check or a quick sign-off does not count as editorial review (Greenberg Traurig).

There are limits. Routine AI-assisted editing, such as a photo touch-up, is not a deepfake. Clearly satirical, artistic or fictional work still needs disclosure, but only in a way that does not spoil the work. Content that was already circulating before 2 August does not have to be labeled retroactively (William Fry).

The Dates and the Fines

Breaches can cost up to €15 million or 3% of worldwide annual turnover (Seyfarth Shaw). More than 180 companies have signed the Commission’s Code of Practice, which is meant to give signatories a practical route to compliance.

The rollout has not been smooth. European broadcasters and advertising groups wrote to the Commission in June arguing that the draft deepfake definition went well beyond what the Act intended and would put labels on content that deceives no one (egta). Software makers raised similar concerns in July (BSA joint statement).

Spain Is Writing Its Own Penalties

The EU rule sets the obligation, but national law decides how it is punished, which means enforcement will not look the same in every country. Spain is moving fastest. On 26 May 2026 the government sent its organic law on the good use and governance of AI to Congress. According to Ara, the bill:

• sets fines from €6,000 up to €35 million for the most serious breaches;
• bans sexual deepfakes outright, while allowing other deepfakes as long as they carry a clear AI label;
• applies to any AI provider serving users in Spain, wherever the company is based;
• gives day-to-day oversight to AESIA, Spain’s AI supervisory agency, with the data protection authority keeping biometric data and the judiciary’s governing council covering AI in the courts.

Ara also notes that until the Spanish law takes effect, the labeling duty applies in Spain but there is no national penalty for ignoring it. Companies operating there should plan for that gap to close.

Why Labels Alone Will Not Protect Your Reputation

The rules bind the people who follow rules. A competitor’s marketing team or a mainstream publisher will add the label. Someone making a fake video of your CEO announcing layoffs, or a cloned voice approving a payment, will not.

That creates an awkward side effect. As audiences get used to seeing “AI-generated” tags on legitimate content, an unlabeled fake can look more believable, not less.

There is a second problem, and it is newer. A fake that circulates for a few days does not just disappear when it is taken down. News stories, social posts and forum threads that discussed it can be indexed and cited by AI assistants such as ChatGPT, Perplexity and Gemini, which then repeat the claim to anyone who asks about the company months later.

A Practical Checklist for Brands and Executives

If someone deepfakes you

1. Verify fast. Confirm the content is fake and preserve copies, URLs and timestamps before anything is removed.
2. Use the new rules in your takedown requests. Point platforms and publishers to the fact that unlabeled synthetic content of a real person now breaches Article 50, alongside defamation, privacy and impersonation grounds.
3. Get your version out first. Publish a short, factual statement on channels you control so journalists and AI systems have an authoritative source to find.
4. Clean up what AI assistants say. Check how the major AI tools describe the incident in the weeks after, and work on the sources they cite until the answers are accurate.

If you use AI yourself

1. Audit your marketing. AI avatars, synthetic spokespeople, cloned voices and realistic AI product imagery featuring people are likely to need a visible label.
2. Label chatbots. Customer service bots must tell users they are AI.
3. Decide who signs off AI-written content. If you want the editorial review exception for public interest text, a named person has to review it properly.
4. Check your vendors before 2 December. Ask the AI tools you use whether they will meet the machine-readable marking duty when the grace period ends.

Getting caught using an unlabeled AI spokesperson is no longer only a compliance issue. It is the kind of story that ends up in the search results and AI answers about your brand.

Conclusion

The EU has made honesty about AI content a legal duty, and the next deadline is weeks away. The rules will not stop bad actors, but they give companies stronger grounds to fight fakes and raise the cost of careless AI use in their own marketing. The brands that come out ahead will be the ones that treat the labels as a starting point and put monitoring, rapid response and AI search cleanup around them.

For companies that want specialist help, see our ranking of the Top AI Reputation Management Companies In 2027.