EU AI Act Article 50: What the AI Labeling Rules Mean for Creators

The EU AI Act's transparency rules for AI-generated content became applicable on 2 August 2026. They mostly bind the companies that build AI systems, but one part applies to anyone who publishes deepfakes. This is a plain summary, not legal advice.

Providers must mark AI output

Article 50(2) requires providers of generative AI systems to mark synthetic audio, images, video and text in a machine-readable format so it can be detected as AI-generated. Fines can reach 15 million euros or 3% of worldwide turnover. The Digital Omnibus on AI, in force since late July 2026, gives systems already on the market before 2 August until 2 December 2026 to comply.

Deployers must disclose deepfakes

Article 50(4) is the part that reaches creators. If you use an AI system to generate or manipulate image, audio or video content that is a deepfake, meaning it resembles real people, places or events and could falsely appear authentic, you must disclose that it was artificially generated or manipulated. For evidently artistic, creative, satirical or fictional work, the disclosure can be made in a way that does not spoil the work.

The Code of Practice

The Commission's Code of Practice on transparency of AI-generated content was finalized on 10 June 2026. The first signatories, published on 31 July, included Google, OpenAI, Meta, Microsoft, Anthropic, Black Forest Labs, Mistral and others, and signing remains open. Its technical section asks providers to:

• use at least two marking layers: signed, timestamped metadata and an imperceptible watermark, with fingerprinting optional;
• offer free detection through a public specification, software or API;
• preserve existing markings and prohibit removing or tampering with them in their terms of service;
• not market or promote tools whose purpose is to circumvent markings;
• make watermark detection interoperable by 2 February 2027.

What is exempt

The Commission's guidelines of 20 July 2026 exempt assistive and standard editing that does not substantially alter content, among other cases. Resizing, format conversion and ordinary retouching are not what the marking duty is about.

What it means in practice

• Expect nearly all images from major AI services to carry both a manifest and a watermark by December 2026.
• Generator terms of service will prohibit removing those markings.
• If you publish realistic AI content that could pass as real, disclose it. Removing metadata does not change that duty.
• Removing personal metadata, such as GPS or device details, from your own photos is unaffected.