Particle.news

EU Requires Labels on Deepfakes as AI Transparency Rules Take Effect

The measure should help users spot synthetic audio, images and video by promoting interoperable watermarking and provenance tools.

Overview

  • The EU's Article 50 labelling obligation takes full effect on August 2 and requires deployers to disclose AI‑generated or AI‑manipulated audio, images and video at first exposure, with existing systems given transitional relief until December 2.
  • The European Commission published final guidance in late July that narrows how the rules apply to games by exempting many clearly fictional uses such as invented voices and fantasy environments while flagging photorealistic recreations of real people or events as likely subject to disclosure.
  • Major tech firms are moving to operationalise the rules: Google signed the voluntary EU Code of Practice and is promoting its SynthID watermark together with the C2PA provenance standard to attach machine‑readable markers and history to media files.
  • Regulators and industry warn that key grey areas remain, including how to treat photorealistic recreations, how to apply the law to content made outside the EU that will be distributed in the bloc, and how to avoid user confusion from overlapping labels.
  • Non‑compliance carries steep penalties of up to €15 million or 3% of global turnover and will raise costs and coordination burdens for smaller creators while changing how developers, advertisers and platforms label trailers, ads and social posts for EU audiences.