Disclosure

Does a deepfake or public-interest text need a disclosure?

Deepfake image, audio, or video, public-interest text, and human editorial control.

What this checks

Deepfake image, audio, or video, public-interest text, and human editorial control.

Who should use this screen

Deployers who publish such output.

Current dates

Article 50 applies from 2026-08-02. Marking transition for earlier systems ends 2026-12-02. Annex III high-risk Chapter III sections 1-3 starts 2027-12-02. Annex I high-risk starts 2028-08-02. Transparency and high-risk stay on separate tracks.

Official sources

Regulation (EU) 2024/1689, Regulation (EU) 2026/1744, Commission Guidelines of 20 July 2026, the Commission FAQ, and the transparency code of practice on EUR-Lex and the Commission site.

Limitations

Decision support only. No high-risk conformity assessment, CE marking, quality-system build, or fundamental-rights workflow is included.

How to read the result

A deepfake stays on disclosure likely, even where a person reviews the text.

Public-interest text with human editorial control is a possible exception, not a clearance.

Other combinations stay on review.

Separate tracks

Transparency dates and high-risk dates are never mixed in the result.

Limits

Does editorial control clear a deepfake?

No. A deepfake remains disclosure likely.

When can an exception appear?

Only for public-interest text with human editorial control, and only as a possible exception.