EU AI labelling rules took effect on Sunday, 2 August 2026, forcing the companies behind generative AI systems to mark deepfakes, synthetic images, cloned voices and machine-written text wherever it could pass as human-made.
The obligations sat inside the European Union’s AI Act, the bloc’s sweeping rulebook for artificial intelligence, and applied to what the law called providers, as reported by The Guardian.
Providers were the firms that built the AI systems, not the people typing prompts into them.
What the EU AI labelling rules actually require
The core demand was a machine-readable marking, a signal baked into the file itself rather than a visible sticker. Think of it as an invisible watermark that a platform, browser or search engine can read automatically, letting it tell a user that an image, clip or paragraph came out of a machine.
Synthetic media was the umbrella term, covering deepfake video of real people, photographs of scenes that never happened, voices cloned from a few seconds of audio, and generated footage.
AI-written text also fell in when it was published to inform the public on a matter of public interest.
Where the EU AI labelling rules stop short
The carve-outs were broad. Personal content made by ordinary users sat outside the rules, as did anything evidently artistic, satirical or fictional, so a comedy deepfake or an AI-generated album cover did not need a badge. The law was aimed at deception, not creativity.
A third exemption mattered more for newsrooms and marketing teams. Where AI-generated material was reviewed and signed off by a human before publication, and that person or organisation took responsibility for it, no label was required at all.
Editorial oversight, in effect, substituted for the marking.
Why the EU AI labelling rules travel beyond Europe
Non-compliance carried weight. Companies that ignored the transparency duty faced fines of up to €15 million or 3% of worldwide annual turnover, whichever figure was larger.
For the largest AI developers, the turnover measure was the one that stung, scaling the penalty to the size of the business.
Readers outside Europe were likely to notice the change regardless. Global platforms rarely build one product for the European Union and another for everyone else, so disclosure labels tuned to Brussels tended to appear on feeds in Johannesburg, London and Los Angeles at the same time.
Enforcement was the next question. Regulators across the 27 member states were handed the job of policing the duty, and the first cases they took up would settle how hard a bar the artistic exemption really set, ahead of the AI Act’s remaining obligations phasing in.
