Since August 2, 2026, the European Union has required AI-generated content to be labeled. The rules come from Article 50 of the EU AI Act, and they’re the first part of the law that most people will actually see: a notice that you’re talking to a chatbot, a disclosure on a deepfake, a hidden mark inside an AI-generated image.
There’s one more date coming. Generative AI systems that were already on the market before August 2 have until December 2, 2026 to add machine-readable marking to what they produce. Here’s what the rules require, who they apply to, and what changes for the rest of us.
What Article 50 requires
The transparency rules split responsibilities between two groups: providers, the companies that build and offer AI systems, and deployers, the businesses and organizations that use those systems.
Providers must:
- Make sure people know when they’re interacting directly with an AI system, such as a chatbot, unless it’s already obvious.
- Mark the output of systems that generate synthetic audio, images, video, or text in a machine-readable way, so that it can be detected as AI-generated or manipulated.
Deployers must:
- Tell people when they’re exposed to emotion recognition or biometric categorization systems.
- Disclose deepfakes: images, audio, or video that have been generated or manipulated to look real.
- Disclose AI-generated text published to inform the public on matters of public interest, unless it has gone through human review or editorial control and someone takes editorial responsibility for it.
The European Commission has published guidelines on how these duties work in practice, and a voluntary Code of Practice on AI-generated content gives companies a recognized way to show they comply.
What happens on December 2
The machine-readable marking duty is the only part with a grace period. Providers whose generative systems were placed on the EU market before August 2 have four months, until December 2, 2026, to comply. Systems launched on or after August 2 had to comply from day one.
Everything else, including chatbot disclosure and the deployer duties around deepfakes, has applied since August 2. According to Cooley’s analysis, content that was generated and published before August 2 doesn’t need to be labeled retroactively. As law firm Morgan Lewis put it in its analysis, the December 2 transition period “does not postpone the deployer duties.”
Who it applies to
The rules apply to providers and deployers whose AI systems are placed on the market or used in the EU, regardless of where the company is based. A US company offering an AI image generator to people in Europe is in scope, just as a European one is.
For violations, Article 99 of the AI Act allows fines of up to €15 million or 3% of a company’s worldwide annual turnover, whichever is higher. For small and medium-sized businesses and startups, the cap is whichever of the two is lower. Enforcement sits mainly with national market surveillance authorities, with the EU AI Office supervising certain systems directly.
What “machine-readable marking” means in practice
The law doesn’t mandate a single technology. In practice, there are two main approaches, and many providers use both:
- Metadata and provenance records attached to a file. The best-known standard is Content Credentials from the C2PA coalition, which records where a piece of content came from and how it was edited, a bit like a nutrition label.
- Watermarks embedded in the content itself, designed to survive some editing and re-saving.
Neither is foolproof. Metadata can be stripped when a file is screenshotted or re-uploaded, and watermarks can be degraded. That’s one reason the rules combine technical marking by providers with visible disclosure by deployers.
What it means for you
If you just use AI tools
You’ll increasingly see labels and notices, especially on images and video. Treat a label as useful information. But don’t treat the absence of one as proof that something is real. Labels can be missing, stripped, or never applied by tools outside the rules’ reach. Our guide on how to fact-check an AI answer includes a short section on checking images.
If your business publishes AI-generated content
Start with a simple inventory of where you use generative AI in anything public: marketing images, product photos, voiceovers, customer-facing chat. Then check two things:
- Does your AI tool mark its output? Many large providers have signed up to the Code of Practice, but don’t assume. Ask your vendor, and make sure your own editing workflow doesn’t strip the marking.
- Do you need a visible disclosure? If you publish realistic AI-generated or AI-altered images, video, or audio of people or events, the deepfake disclosure likely applies. If you publish AI-generated text on matters of public interest without real editorial review, that disclosure may apply too.
When in doubt, disclose. A short “created with AI” note costs you very little, and it’s the direction regulation and reader expectations are both moving.
If you build an AI product
If your system generates content and people in the EU can use it, the marking duty applies to you, and if you launched before August 2, December 2 is your deadline. The Commission’s guidelines and the Code of Practice are the best places to start.
My take
I think labeling rules are a reasonable floor, not a solution. Marking is imperfect, and bad actors will ignore it. But for the vast majority of businesses that use AI honestly, these rules mostly formalize what’s already good practice: tell people when they’re talking to a machine, and don’t pass off synthetic media as real.
This article is a general explanation, not legal advice. If your business is in scope, talk to a lawyer about your specific situation.
Sources
- European Commission, Guidelines on transparency obligations for providers and deployers of certain AI systems
- European Commission AI Act Service Desk, Article 99: Penalties
- Morgan Lewis, EU AI Act’s transparency rules: what went into effect on 2 August, August 2026
- Cooley, EU AI Act: transparency obligations take effect 2 August 2026, August 2026
- C2PA, the Coalition for Content Provenance and Authenticity
