Source: The Guardian

Synthetic text, images, video and audio that are designed to appear truthful must be visibly marked as AI-generated and include a digital watermark, under rules developed as part of the EU’s AI Act; texts on matters of public interest must also be labelled when there is no human editorial oversight. New AI systems on the EU market must comply from 2 August, with existing systems given an extra four months; users’ personal content is exempt and there is an exemption for “evidently artistic”, satirical and fictional works, while companies are encouraged (but not required) to label content created before the rules. The executive has produced black-and-white labels as a baseline, although providers may design their own labels.

Failure to comply risks fines of up to €15m or 3% of worldwide global turnover. The reporting (from Brussels) records industry concern that EU guidance has broadened the scope of what must be labelled, risking widespread “label fatigue” and burdens on sectors such as advertising, film and publishing, a point raised by CCIA Europe. The piece cites examples of deceptive deepfakes reported in recent years — audio purporting to show Slovak opposition leader Michal Šimečka conspiring, doctored video of Emmanuel Macron, and fake footage of Keir Starmer — and notes tech measures already in use, including TikTok’s claim to have labelled more than 3bn items and Google’s SynthID watermarking of billions of images and extensive audio. The article records Meta’s stated intention to sign the EU code of practice (announced the week of 28 July) and quotes EU lawmakers and officials arguing the transparency rules protect both consumers and democracy while the commission prepares broader enforcement under the AI Act.