Published 2026-09-15 · Updated 2026-11-08
AI Content Labels in 2026: EU Article 50, California AB 723 and Platform Rules
Before publishing AI content, separate three checks: the law that applies to your role and market, the platform’s disclosure rules, and whether the file still contains provenance information. A watermark or a caption alone is not proof of compliance. This guide is practical information, not an assessment of your individual legal obligations.
EU: providers and creators have different duties
Article 50 distinguishes technical marking by providers from disclosure by deployers of deepfake content. Standard assistive editing and artistic or fictional works have specific qualifications; fictional work is not a blanket exemption from disclosure. Territorial scope depends on the operator, use and EU connection, not simply the presence of one EU follower.
The Commission’s current timeline applies transparency obligations from 2 August 2026. Providers of systems placed on the market before that date have until 2 December 2026 for Article 50(2) marking and detection. Check the current Commission guidance and your role before applying a deadline.
California: show the change and link the original
AB 723, effective 1 January 2026, covers digitally altered property-sale advertising by brokers, salespersons and people acting for them. Put a reasonably conspicuous disclosure on or adjacent to the altered image and a public URL, link or QR code clearly identifying the unaltered original. A controlled website must include the original or a qualifying public link. Keeping it available only on request is insufficient.
Ordinary photo adjustments that do not change the property’s representation are excluded. Preserve the original capture and label virtual staging or renovation concepts; do not present a concept as an existing property feature.
Platforms: use the disclosure control
TikTok requires labels for realistic AI-generated images, audio and video. YouTube asks creators to disclose meaningfully altered or synthetic realistic content; minor edits and clearly unrealistic content are treated differently. Meta uses AI info labels based on industry signals or user disclosure. Check the relevant upload flow rather than assuming metadata will do everything.
Provenance: keep originals and inspect exports
SynthID is an embedded watermark; C2PA records provenance credentials. They are different mechanisms, and normal editing can affect them differently. Keep original outputs, references and consent records. Check the final export and add any required visible disclosure; the presence of a credential does not itself establish legal compliance.
A repeatable publishing checklist
Record which parts were generated or materially edited. Confirm rights and consent for reference faces and voices. Preserve original photos and outputs. Add the appropriate platform disclosure and any legally required statement or original-image link. Review the published post to confirm the label and links are visible.
FAQ
Must every AI-edited image carry the same label?
No. The applicable role, type of alteration, jurisdiction and platform matter. Distinguish minor assistive edits from realistic synthetic or substantially altered content.
Is the original listing photo available only on request enough in California?
No. For advertising covered by section 10140.8, provide the required disclosure and publicly accessible original through the prescribed link, URL or QR code.
Does C2PA or SynthID replace visible disclosure?
No. Provenance signals and audience-facing disclosure have different purposes. Keep provenance and apply the disclosure requirements that cover your publication.