EU mandates AI labels on realistic fake content from Sunday

From August 2, 2026, companies must visibly label AI-generated images, audio, and text that appear authentic and include a digital watermark, under EU rules.

The AI Act provision targets deep fakes that have deceived voters and smeared officials.

New AI systems entering the EU market face the mandate immediately; existing systems have until December 2, 2026, to comply.

Texts on public interest matters also require a label, unless a human editor has reviewed them.

Personal content and works clearly artistic, satirical, or fictional are exempt. Companies are encouraged but not required to label older AI-generated material.

Non-compliance risks fines of up to €15 million or 3% of a company’s global annual turnover.

The European Commission offers black-and-white labels, but firms may design their own.

Industry supports the goal but fears overbroad interpretation. The Computer and Communications Industry Association argues that guidelines from July 2026 expanded the deep fake definition beyond the original act.

Boniface de Champris, CCIA Europe’s AI policy lead, said the label was meant for deceptive content, but now almost everything gets labeled, which could desensitize users like cookie banners. He expects the biggest impact in advertising, film and publishing, sectors where AI is common but unnoticed.

TikTok already requires creators to label realistic AI content and has helped label over 3 billion pieces. Google’s SynthID tool, launched in 2023, has watermarked more than 100 billion images and 60,000 years of audio. Google and Meta have signed a voluntary EU code of practice, joined by over 180 organizations.

On July 28, 2026, Markus Reinisch, Meta’s vice president for public policy in Europe, said the company would sign the code to prevent a confusing patchwork of labels.

Green MEP Sergey Lagodinsky, who helped negotiate the act, expressed confidence industry could adapt, noting that past complaints often diminished after implementation.

The labeling rules are part of broader AI Act enforcement. Henna Virkkunen, the Commission’s tech policy lead, said the framework gives innovators legal certainty while safeguarding the public interest.