EU's New AI Transparency Rules Now in Force: What Gets Labeled, Who Pays, and When
Published on 08/08/2026 at 19:32 | Redaktion boerse-global.de
The clock has run out on the grace period. Since August 2, 2026, Article 50 of the EU AI Act has been binding law across the bloc, requiring anyone deploying generative tools to clearly mark synthetic content. The stakes are steep: violations can trigger fines up to €15 million or 3 percent of global annual turnover, whichever is higher.
The Labeling Mandate, Explained
The transparency obligations sweep broadly. Deepfakes, AI-generated imagery, video, and audio all fall under the new regime. So do text outputs dealing with matters of public interest—provided they are predominantly machine-produced and never passed through human editorial review.
Individual users are not exempt simply because they are not corporations. The rules apply to everyday people too, though purely private use stays outside the scope. Carve-outs also exist for legally sanctioned law enforcement activities, artistic works, and any content that has undergone genuine human editorial oversight. Meanwhile, developers of generative systems carry a separate duty: they must ship technical mechanisms that allow machine-readable labeling.
For systems already on the market, the law offers breathing room. Existing products must comply by December 2, 2026. High-risk systems get until December 2, 2027. Regulatory sandboxes—testbeds for AI oversight—were supposed to be operational by the August 2 deadline.
Icons, Watermarks, and Industry Moves
To smooth implementation, the European Commission published official labeling icons on August 6, 2026. The pictograms come in four variants, available in both SVG and PNG formats, and are free to use. They serve as the standardized visual marker under Article 50(4).
Corporate responses have been quick. WhatsApp is rolling out a feature that flags AI-generated images and videos under specific conditions. Music platform Suno announced on August 7, 2026 that it will embed watermarks into AI-composed tracks, with the technology slated for rollout in the coming weeks. Suno's mandatory machine-readable marking must be live by December 2, 2026, per EU requirements.
New compliance deadlines like this one can catch businesses off guard, and workplace safety rules come with their own ticking clocks. A free toolkit with 41 ready-to-use checklists helps you document risks properly and stay ahead of enforcement action. Download the free Risk Assessment Toolkit
The music sector is also navigating training-data disputes. Labels including Sony and Warner have already exercised their opt-out rights to prevent their catalogs from being used in AI training. Legal friction is mounting: the Munich Regional Court recently ruled that Suno infringed GEMA rights, adding another layer to the ongoing tension between creators and AI developers.
Doubts Among Experts, Nuance at Broadcasters
Not everyone is convinced the labeling regime will deliver on its promise. Maria Pawelec of the University of Tübingen cautioned that marking AI content alone will not resolve the deepfake problem—and could paradoxically deepen public distrust of all digital media. Other researchers have echoed similar reservations, warning that over-labeling may breed cynicism rather than clarity.
Just as regulators demand clarity on synthetic content, they also expect businesses to keep their safety documentation in order. Over 37,000 UK companies rely on a free Health & Safety Toolkit with ready-made risk assessments and checklists covering key regulations like COSHH and PUWER. Get the free Health & Safety Toolkit
Broadcasters are already experimenting with layered approaches. Bavarian public broadcaster BR employs a three-tier labeling system: the highest tier marks realistic AI content, the middle tier covers artistic depictions, and the lowest tier flags non-realistic elements. The station also operates on a "human-in-the-loop" principle, meaning no piece goes live without human sign-off.
