The European Union sets a date for the end of unlabeled AI content
As of 2 August 2026, any content generated wholly or partly by artificial intelligence that circulates in the European Union —whether text, image, video or audio— must carry a clear label indicating its synthetic origin. This is established by the Artificial Intelligence Regulation (AI Act) approved in 2024, whose first binding provisions come into force this summer. The measure affects platforms such as Facebook, TikTok, YouTube, X and Google, but also any website, media outlet or content creator that publishes from or for the EU.
What exactly changes
The obligation is twofold. On the one hand, generative AI systems such as ChatGPT, Gemini, Midjourney or video generators like Sora will have to technically mark their creations with machine-readable metadata. On the other hand, any platform that distributes AI-generated content will have to display to the user a visible label or notice indicating that what they are seeing is not real. This ranges from hyperrealistic deepfakes to simple automatic text summaries.
The European Commission will be in charge of supervising compliance and fines can reach 7% of the global annual turnover of the offending company. To give you an idea, in the case of Meta (parent company of Facebook and Instagram), that would be more than 9 billion euros.
Why this law is necessary
The proliferation of generative artificial intelligence has caused cases of disinformation to soar. According to a report by the European Digital Media Observatory, political deepfakes multiplied by five during 2025, coinciding with key electoral processes in several European countries. AI-generated disinformation is no longer limited to poorly written texts: today there are hyperrealistic videos of politicians saying things they never said, cloned audio of relatives asking for money, and photographs of events that never happened.
The problem is that AI has reached a point where the human eye can no longer distinguish the real from the synthetic. Recent studies show that, even with specific training, people get it right only in 55% of cases when trying to identify whether an image was generated by AI.
What happens with memes, digital art and creative uses
The law provides for exemptions for artistic, satirical or parodic content, as long as there is no intention to deceive the public. A meme created with AI will remain legal, but if it is spread as if it were real news, it falls within the regulation. The key lies in intentionality and in whether the content could mislead a reasonable person.
How it will be applied in practice
The big platforms have been preparing for months. Meta has already announced that it will automatically label images uploaded to Facebook and Instagram that it detects as AI-generated, using C2PA metadata. Google has integrated its SynthID system into search results and YouTube. OpenAI added visible and invisible watermarks to DALL-E 3 images.
To address fragmentation, the European Commission is developing a voluntary certification system that will allow any creator to demonstrate that their content complies with the regulation. In addition, a public repository of synthetic content detection algorithms will be created.
What this means for the average user
You will start seeing labels such as “AI-generated content” or “synthetic audio” in more places. When you share an image created with an AI tool, the platform will ask you to label it. And, above all, you will gain a tool you don’t have today: the ability to know, with some legal backing, whether what you are seeing is real or fabricated.
It is not a magic solution against disinformation, but it is the world’s first legal framework that assigns duties to artificial intelligence instead of merely observing it. And it sets a precedent that other countries are already studying to copy.
Marta, for inteligencia intermitente.
