Europe's New AI Disclosure Rules

New rules took effect across the European Union on August 2, and the idea behind them is simple: if you are using AI to communicate with people, you are supposed to tell them. The EU AI Act is the world's first comprehensive AI law, phasing in since 2024, and the phase that just landed covers how you communicate with your customers. That is Article 50, and it applies to any AI output used or accessed in the EU, regardless of where it was created or who it was aimed at

What happened

This is not an isolated compliance update. It sets a global floor for AI disclosure that reaches any brand touching European audiences, and it lands squarely on communications workflows: chatbots, synthetic media, and published content on matters of public interest. That is it. Everything else is detail about where the line falls.

• If an AI system talks directly to people, they have to be told they are dealing with AI. A line in your terms and conditions, a watermark, calling it an "assistant," or saying it is "powered by large language models" does not count. What does: a plainly worded notice where the conversation happens, and an audio cue for voice.

• You have to disclose AI-generated images, video and voice that could pass for real. Europe's definition is wider than most US rules. It covers objects, places and events, not just people, and it does not require a real person or an intent to deceive.

• AI-written text needs disclosure when it is published to inform the public on matters of public interest, such as health, the environment, consumer protection, or economic and political developments, unless a human genuinely reviewed it and a person or organization takes editorial responsibility for it.

• What stays outside: animation, illustration and anything obviously unreal; a real product against an AI background, as long as the ad does not mislead about the product; and routine post-production like color correction and audio cleanup.

Practical moves

When in doubt, disclose. The rule draws fine lines about what counts as realistic enough and which market you were aiming at, and disclosing means you never have to argue either one.

Be specific in your disclosures. The rule wants disclosure that is clear and not misleading, and specificity is how you get there. "Made with AI" means almost nothing, and on a real product shot with an AI background it can even mislead by implying the product itself is fake. "The background in this image is AI-generated" is both clearer and more accurate.

Match the disclosure to the format. Nothing says the disclosure has to be prominent, only that it is clear and someone meets it on first exposure. For video, an end card, a line of fine print, or a note in the description. For social, the caption. For images, a credit line. For audio, say it in the intro or put it in the show notes.

Update your AI policy. Most policies cover data and tooling and stop there. Write the disclosure standard into yours.

Name an editorial owner wherever one does not exist. The exposure is content on real subjects produced faster than any human is reading it closely: automated news feeds and industry roundups, AI-drafted sustainability reports, health or safety information generated at scale, and investor updates.

Put synthetic media in your brand guidelines. Almost nobody has done this. Your guidelines already govern logo, color, tone and photography. Add a position on AI likenesses, voice cloning, what you will never generate, and how it all gets disclosed. It turns a hundred scattered judgment calls into one decision made in advance.

Use the self-flagging tools you already have. Most social platforms now let you flag AI content yourself when you post. Use it. It takes one click, it puts the disclosure where people see it, and it means you are not relying on a platform to detect something you already know.

Looking ahead

Google signed the EU's transparency code in July and is watermarking generated content. YouTube can add its own label when a creator doesn't disclose, especially for realistic or sensitive content like elections, health, and finance. These platform moves reduce friction, but they don't transfer accountability. Responsibility still sits with you. If you use an off-the-shelf bot, the vendor has to give you the capability to disclose, but you are the deployer, so it is on you to turn it on and word it properly.

If labeling it would embarrass you, do not make it. That test will get you further than anything in the regulation

Go deeper: Article 50 | The Commission's guidelines

Phyusion is not a law firm. For guidance specific to your situation, talk to counsel.

Connect

We're here to listen, collaborate, innovate.

What's on your mind?

Privacy Policy

© 2024. All rights reserved.