The EU’s New AI Labeling Rules Just Took Effect: What They Mean

What are the EU AI Act’s new transparency rules?

The EU AI Act’s transparency rules are a set of disclosure duties, in Article 50 of the law, that force companies to be open about when and how they use AI. They took effect on August 2, 2026, and the goal, per the European Commission, is to help people recognize when they are dealing with an AI system or looking at AI-generated content. In short, the machine has to say it is a machine.

The rules sit inside Regulation (EU) 2024/1689, the broader AI Act, and the European Commission published its final guidelines interpreting Article 50 on July 20, 2026. You can read the plain-language summary on the European Commission’s official transparency page. The duties fall on two groups: providers, which build AI systems, and deployers, which put them to use.

This is a narrow, specific obligation rather than a ban on anything. No AI tool is outlawed here. The requirement is disclosure, so people are not misled into thinking AI output is human-made or that a synthetic video is real footage.

What exactly has to be labeled or disclosed now?

Article 50 covers four situations, and the type of disclosure depends on the situation. Each one targets a case where a person could reasonably be fooled about whether AI is involved. Here is the breakdown.

There are real exceptions. Clearly artistic or satirical work gets lighter treatment, and obviously fantastical content, such as a person flying unaided, does not count as a deepfake, per the Commission’s official Article 50 guidelines. Law-enforcement uses for detecting or investigating crime are also exempt. So the rules aim at deception, not at every playful AI edit.

Does the EU AI Act affect you in the US?

The EU AI Act can affect you in the US indirectly, even though it is a European law. It applies to any provider or deployer putting AI in front of people in the EU, regardless of where the company is based, so a US firm serving European users is in scope. Many large AI products are global, so companies often apply one consistent approach rather than build a separate US version.

That means US users may start seeing AI disclosures and content labels roll out broadly, simply because the providers behind those tools operate in Europe too. You are not protected by the EU law as an American, but you may benefit from the labels it prompts. The practical effect depends on whether each company localizes its compliance or applies it worldwide.

For businesses, the reach is the headline. A US company that runs an AI chatbot for EU customers, or publishes AI-generated marketing to European audiences, carries these duties even if the AI tools belong to a third party. For context on how US regulators approach AI honesty, see our explainer on the FTC’s AI accuracy policy.

Will you actually start seeing AI labels on content?

You will likely see more AI labels over the coming months, but the rollout is gradual rather than instant. The disclosure duty for chatbots and deepfakes applies now, while the machine-readable marking of AI-generated content has a grace period until December 2026 for tools already on the market before August 2, per the Commission. Content made before August 2 does not need retroactive labeling.

Two forms of labeling matter here. One is visible, such as a chatbot saying it is AI or a video carrying an on-screen note. The other is invisible, a machine-readable marker embedded in a file so platforms and tools can detect AI origin automatically. The invisible kind is what enables detection at scale, and it is the piece still being standardized.

Enforcement gives the rules teeth. Non-compliance can bring fines up to 15 million euros, about $16 million, or 3% of a company’s worldwide annual revenue, whichever is higher, with a lower ceiling for small businesses and startups. That scale is why global AI firms are paying attention rather than ignoring a foreign rule.

What does this mean for deepfakes and scams?

For deepfakes, the rules require disclosure, but they do not make fraud disappear, so healthy skepticism still matters. A criminal running a voice-clone scam or a fake video of a CEO is already breaking the law and will not politely add a label. The disclosure duty targets legitimate deployers, which helps set a norm, not the bad actors who cause the most harm.

Stop here and verify through a separate, trusted channel if a video or voice message pressures you to send money or share a code, because a convincing fake will not carry a warning. That habit protects you regardless of any labeling law. Governments are also acting on the worst abuses, as we covered when governments moved against non-consensual AI imagery.

The upside for everyday users is a slow shift toward being able to tell synthetic from real. As detectable markers spread across AI image tools and AI video generators, platforms can flag AI content more reliably. That will not catch determined fakers, but it raises the floor for honest content.

How does this compare to US rules?

Compared with the US, the EU is far ahead on binding AI-transparency law, and that gap is the main story for American readers. The US has no equivalent federal statute requiring AI disclosure or content labeling as of August 2026. Federal action so far has leaned on existing tools, such as the Federal Trade Commission policing deceptive AI claims, rather than a dedicated labeling mandate.

Several US states have moved on pieces of the problem, particularly election-related deepfakes and, in California, some AI transparency requirements. Coverage of these state laws varies, and legal challenges are ongoing, so the US picture is a patchwork rather than a single rule. Supporters of the EU approach argue clear labeling builds trust, while critics say the duties are broad and costly to apply, especially for smaller firms. Both concerns are being tested now that the rules are live.

For anyone comparing the AI tools themselves rather than the laws, disclosure duties will increasingly shape how chatbots present themselves. Our roundup of Claude vs ChatGPT vs Gemini looks at those assistants, which now sit squarely inside the direct-interaction rule when they serve EU users.

FAQ

When did the EU AI Act transparency rules take effect?

The Article 50 transparency obligations took effect on August 2, 2026. They apply to AI systems placed on the EU market regardless of when the system was first released.

Do US companies have to follow the EU AI Act?

Yes, if they provide or deploy AI systems used by people in the EU, US companies fall under Article 50 regardless of where they are based. A US business running an AI chatbot or publishing AI content for EU audiences is in scope.

Does AI-generated content have to be labeled in the US?

No, the US has no federal law requiring AI content labeling as of August 2026, though some states regulate deepfakes and AI transparency. US users may still see labels because global companies often apply EU rules broadly.

What are the penalties for breaking these rules?

Fines can reach 15 million euros, about $16 million, or 3% of worldwide annual revenue, whichever is higher. A lower ceiling applies to small businesses and startups.

Does every AI-generated image now need a label?

Not indiscriminately, since the rules focus on cases that could mislead people, with exceptions for clearly artistic, satirical, or fantastical content. Providers must add machine-readable markers to synthetic content, subject to a grace period into December 2026 for existing tools.