EU AI Act: AI Transparency Rules Take Effect — and What Remains Unresolved

On 2 August 2026, the AI content transparency requirements under Article 50 of the EU AI Act formally took effect in the European Union. It is the first mandatory standard for labeling synthetic content in the world's largest digital market. The Code of Practice (CoP) for generative AI providers took effect at the same time.

What providers must now do

Under the Code of Practice, providers of generative AI systems must meet four key requirements as of 2 August 2026:

• Multi-layer labeling — embed digital metadata and watermarks in the content they generate
• Free detection tools — give users AI-content detectors that comply with EU privacy law
• Quality standards — ensure labeling systems are robust and interoperable
• Compliance processes — test regularly and demonstrate that requirements are met

Deployers of AI systems must tell users when they are interacting with AI, according to Section 2 of the Code. It is the first large-scale mechanism that gives audiences the right to know where content comes from.

Why experts call the rules incomplete

The rules' key weakness is their broad exemptions. The EU AI Act does not require labeling for minor edits (cropping, framing, AI translation), content on matters of public interest, or material produced for personal non-commercial purposes. The personal-use exemption is the one critics call most dangerous: it puts privately made viral deepfakes outside the regulation, and those are the ones that do the most harm.

"A label is not protection, and the two should not be confused," note the authors of a

Tech Policy Press report published to coincide with the rules taking effect.

Provenance data remains optional, so it is still impossible to trace a synthetic item's full path through the distribution chain. Another systemic problem is open-weight models: voluntary compliance stops mattering once a model has been released and fine-tuned by third parties. Public reporting and independent oversight by civil society also remain limited.

How reliable is AI-content detection?

AI-content detectors are not yet a dependable line of defense. According to TRIED, a benchmark built specifically to evaluate synthetic-media detection systems, results remain unreliable and uneven, with quality varying significantly by language and by the subject being filmed. The authors warn explicitly that detectors cannot be treated as a reliable last line of defense.

Watermarks and metadata are easily lost when AI content is edited and re-shared. Without robust provenance mechanisms, labeling becomes a symbolic gesture.

The trend is not limited to the EU: the US states of California and New York are introducing similar measures in 2026, drawing on the principles of the Coalition for Content Provenance and Authenticity (C2PA). Still, no jurisdiction has yet solved the problem of synthetic content systematically.

What this means

The EU AI Act has taken the first mandatory step toward AI content transparency, but critical loopholes remain. Optional provenance, the personal-use exemption and unreliable detectors mean that real protection is still to come. The key question is whether regulators will close these gaps or wait for the industry to sort things out on its own.

FAQ

What is the EU AI Act Code of Practice?

The Code of Practice (CoP) is a mandatory set of rules for generative AI providers in the EU. It took effect on 2 August 2026, together with the Article 50 requirements. It obliges platforms to implement content labeling, provide free detectors and undergo regular compliance checks.

What content is exempt from labeling?

Exempt are minor edits (cropping, framing, AI translation), content on matters of public interest, material for personal non-commercial purposes, and content generated in real time where labeling is technically impossible. Experts criticize the personal-use exemption as the most dangerous when it comes to the spread of deepfakes.