AIBRIEF
The AI Act's first rule takes effect for new AI systems on the EU market, with up to 3% of gross revenue as the penalty for non compliance.
On August 2, any AI-generated image, audio clip, or text designed to look authentic and placed on the EU market has to carry a digital "made by AI" label. The most surprising place readers will see the change is not on social media but in the ad break, the film cut, or the marketing email they assumed was human-made.
The rule, the first to take effect under the EU's flagship AI Act, applies to new AI systems released on the EU market from that date. Pre-existing systems get an extra four months. The compliance mechanism is a digital watermark; the EU provides black-and-white labels any organization can adopt, but custom labels are also permitted. Personal and non-commercial content is exempt, as is "evidently artistic" satire and fiction.
The penalty for non-compliance is up to 3% of a company's total gross revenue.
Sergey Lagodinsky, a European Parliament member and AI Act negotiator, framed the rule as both consumer and democracy protection: labels tell buyers what they are looking at, and they protect public discourse from synthetic content that passes as authentic. Boniface de Champris, AI policy lead at the Computer & Communications Industry Association, expects consumer surprise in advertising, film, and publishing, where AI use is widespread but often invisible.
Platforms already label some content. TikTok requires creator-side AI labels, though enforcement is uneven. Google's SynthID says it has marked more than 100 billion images and roughly 60,000 years of audio; Meta labels photorealistic AI images. Both figures are platform self-reports, not independent measurements.
For most readers, the rule's effect will be patchy. The same AI-generated content can carry a label in a paid ad but not in a private message.