The same idea, drafted differently elsewhere

Synthetic content is one of the few areas where several jurisdictions have converged on a similar structure at a similar time. The structures rhyme, which makes it tempting to assume one compliance approach serves everywhere. The details differ enough to matter.

China

The Measures for Labelling AI-Generated Content, together with the mandatory national standard GB 45438-2025, have been in force since 1 September 2025. They require both an explicit visible label and an implicit machine-readable label carried in metadata, and the metadata label must contain the provider's name and a content reference number.

Two things are worth drawing out. First, the same two-part structure appears as in the EU: one label for people, one for machines. That is not coincidence; it reflects that the two audiences genuinely need different things. Second, the Chinese standard specifies what the machine-readable label must contain. A named provider and a reference number make the marking traceable to a source rather than merely flagging that something was generated. It is a stricter specification of the same idea.

If your organisation publishes into the Chinese market, this applies on its own terms, and compliance with Article 50 does not establish compliance with GB 45438-2025.

The EU, restated as a sequence of dates

Article 50 applies from 2 August 2026. It was not deferred by the Digital Omnibus, Regulation (EU) 2026/1744, which was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. That instrument deferred standalone Annex III high-risk obligations to 2 December 2027 and embedded Annex I obligations to 2 August 2028, and left prohibitions, general-purpose AI obligations and Article 50 alone.

The marking grace period for systems placed on the market before 2 August 2026 ends 2 December 2026.

Guidance exists. The Code of Practice on Transparency of AI-Generated Content has been confirmed adequate, and the Commission published guidelines on Article 50 on 20 July 2026. Those guidelines are where the requirement for a visible or audible label understandable without a detection tool comes from.

Two related prohibitions, on the same date

The Digital Omnibus added two prohibited practices under Article 5 which apply from 2 December 2026: AI systems that generate or manipulate non-consensual intimate imagery of identifiable people, including so-called nudifier applications, and AI systems that generate child sexual abuse material. The prohibition reaches systems where such output is a reasonably foreseeable and reproducible result in the absence of effective safeguards.

This belongs in a synthetic content module because it marks the boundary of the subject. Most of Article 50 is about telling people what they are looking at. Article 5 is about content that may not be produced at all, whatever is disclosed. Penalties reflect that: Article 99 sets up to 35 million euro or 7 per cent of global annual turnover for prohibited practices, against 15 million euro or 3 per cent for most other breaches.

The United States, where the picture is fragmentary

There is no single federal labelling rule to point at. California has enacted transparency duties in adjacent areas: AB 2013, operative 1 January 2026 and reaching back to systems released on or after 1 January 2022, requires twelve enumerated training-data disclosures to be published on the developer's website. SB 53, effective 1 January 2026, requires a published Frontier AI Framework and transparency reports at deployment from large frontier developers, with critical safety incident reporting to Cal OES within 15 days, or 24 hours where there is an imminent risk of death or serious physical injury.

Neither is a synthetic content labelling regime. They are worth knowing because they are the instruments people cite when asked what the United States requires, and answering the wrong question confidently is how a compliance register ends up with a gap in it.

The practical consequence

If you publish generated content across markets, you need a record, per piece, of what was generated, by which system, what visible label was applied, what machine-readable marking the provider applied, and where it was published. That record is what lets you answer any of these regimes without re-investigating. Nobody builds it retrospectively.

As of 24 September 2026.