Wiki · Governance · Last reviewed June 25, 2026

AI-Generated Content Transparency Code

The EU Code of Practice on Transparency of AI-Generated Content is a voluntary compliance framework for Article 50 duties on marking, detecting, and labelling certain AI-generated or manipulated content.

Category: Governance Updated: June 25, 2026 Tags: EU AI Act, Article 50, synthetic media, disclosure, provenance

Definition

The Code of Practice on Transparency of AI-Generated Content is an EU AI Office-facilitated code for implementing parts of Article 50 of the EU AI Act. The European Commission published it on June 10, 2026. It is voluntary, but the underlying Article 50 transparency requirements are legal obligations where they apply.

The code addresses a narrow problem: how providers and deployers should make certain AI-generated or manipulated content identifiable. The Commission says Article 50 obligations apply from August 2, 2026 and cover marking and detection of AI-generated content, plus labelling of deepfakes and certain AI-generated publications. The FAQ adds a transitional period until December 2, 2026 for AI systems placed on the market before August 2, 2026.

This entry treats the code as governance infrastructure for Synthetic Media and Deepfakes, Content Provenance and Watermarking, and AI Slop. It is not a proof of truth, authenticity, safety, or lawful use.

Structure

The Commission describes two sections. Section 1 is for providers of generative AI systems under Article 50(2). It concerns machine-readable marking and detection of AI-generated or manipulated audio, image, video, and text, with technical solutions expected to be effective, interoperable, robust, and reliable as far as technically feasible.

Section 2 is for deployers under Article 50(4). It concerns labelling deepfakes and AI-generated or manipulated text published to inform the public on matters of public interest. The Commission's FAQ says this section gives guidance on label design, placement, and presentation, while considering artistic, creative, satirical, fictional, and analogous works, plus human review and editorial responsibility.

The EU also published optional icons for labelling AI-generated content. The icon page says the icons may help deployers disclose deepfakes and relevant AI-generated or manipulated text, but also says icon use alone does not establish legal compliance. That distinction matters: a label is evidence of a disclosure attempt, not evidence that the underlying claim is true.

Agent Context

Agent systems complicate Article 50 because agents can create, edit, post, summarize, translate, or remix media on behalf of users. A browser agent that generates a public-interest explainer, a newsroom agent that drafts captions, or a commerce agent that rewrites product images may create material that enters labelling workflows.

Audit logs should separate who prompted the system, which model or tool generated or modified the content, whether a human reviewed it, where it was published, and whether machine-readable marking or visible labelling survived export, upload, recompression, screenshotting, or reposting.

Governance Use

A deployment record should name the role: provider, deployer, platform, publisher, editor, agency, model vendor, or tool integrator. It should identify the content type, Article 50 route, marking method, visible label, icon use, metadata fields, detection method, accessibility treatment, export behavior, and exceptions claimed.

For provider-side marking, the evidence should say whether outputs are marked by default, whether the mark is machine-readable, what standards or technical methods are used, and what transformations break the signal. For deployer-side labelling, the record should show the first-exposure label, placement, language, icon or disclaimer, and whether a second information layer is available.

The signatory process is also governance evidence. The Commission's signing instructions say providers and deployers may sign the relevant chapter by submitting a form, and that signatories will be publicly listed in July 2026 before the Article 50 date. The Commission and AI Board were still assessing code adequacy after publication.

Limits

The code does not replace the AI Act or Commission guidelines. It does not cover every Article 50 issue, every jurisdiction, every platform rule, or every form of manipulation. It supports specific marking, detection, and labelling obligations.

Labels can fail culturally even when they work technically. They can be too small, too late, too ambiguous, stripped in reposting, hidden by interface overlays, ignored by audiences, or misread as proof that labelled content is false and unlabelled content is authentic.

Machine-readable marking has parallel limits. Metadata can be stripped, watermarks can be attacked, detectors can be unavailable or wrong, and provenance can show origin without proving truth, consent, editorial quality, or legal compliance.

Review Record

Source Discipline

Use Commission pages for claims about the code, FAQ, icons, signing process, adequacy assessment, and forthcoming guidelines. Use EUR-Lex or the AI Act Service Desk for Article 50 text. Treat law-firm summaries, vendor posts, and press commentary as secondary explanation, not as authority for compliance.

Spiralist Reading

Spiralism reads the transparency code as a label discipline for the synthetic public world. It cannot make an image honest, a caption accurate, or a platform fair. Its value is humbler: it gives institutions fewer excuses for letting generated surfaces float without source, method, or responsibility.

Sources

Return to Wiki