Direct answer: Meta announced on July 28, 2026 that it will sign the EU AI Act Code of Practice on Transparency of AI-Generated Content. The underlying Article 50 transparency obligations apply from August 2, 2026. The code itself is voluntary, but the European Commission says the Article 50 requirements are legal obligations. Providers and deployers should now identify which AI outputs need machine-readable marking or visible disclosure, assign owners, test labels, and retain compliance evidence.
Important: This is a practical implementation guide, not legal advice. Confirm your exact role, content type, exemptions, and jurisdiction with qualified counsel and the European Commission’s final guidance.
What Meta announced
In its official announcement, Meta said it is signing the EU code focused on transparency for AI-generated content. Meta connected the decision to its existing work on identifying and labeling AI-generated media across its platforms and to cross-industry provenance efforts such as C2PA.
Meta also said transparency measures should be practical, interoperable, and clear rather than creating a confusing collection of incompatible labels. The announcement does not remove the need for each business to assess its own obligations.
August 2 deadline: what the EU says
The European Commission’s official code page says the Article 50 transparency obligations apply from August 2, 2026. It divides the code into two practical sections:
- Providers: rules for marking and detection of AI-generated or manipulated content.
- Deployers: rules for labeling deepfakes and certain AI-generated or manipulated text.
The Commission describes adherence to the code as voluntary. However, it says Article 50’s transparency requirements remain legal obligations. Signatories can use the code’s measures to demonstrate compliance; organizations using another approach must be able to show that their measures are adequate.
Provider vs. deployer: quick decision table
| Question | Likely track to investigate | Official focus |
|---|---|---|
| Do you provide a generative AI system that creates audio, images, video, or text? | Provider | Machine-readable marking and detectability of generated or manipulated output. |
| Do you use an AI system to publish or distribute a deepfake? | Deployer | Disclosure that the content is artificially generated or manipulated. |
| Do you publish AI-generated or manipulated text about matters of public interest? | Deployer | Disclosure may be required, subject to the scope and exceptions in Article 50 and the final guidance. |
| Does the publication undergo human review and carry editorial responsibility? | Review the exception carefully | The Commission notes an exception for certain public-interest text that has undergone human review and is subject to editorial responsibility. |
Use the table only as a triage tool. Read Article 50 and the Commission’s transparency guidelines before making a final classification.
9-step AI-content transparency checklist
1. Map every AI-content workflow
List the models, APIs, creative tools, publishing systems, and vendors that generate or manipulate text, images, audio, and video. Include automated social publishing, ad creative, customer support, synthetic presenters, and localization workflows.
2. Assign provider and deployer roles
Document whether your organization provides an AI system, deploys one, or does both in different workflows. Assign an accountable owner for each use case instead of treating “AI compliance” as one generic task.
3. Classify content by disclosure risk
Flag deepfakes, realistic synthetic media, manipulated recordings, and AI-generated public-interest publications for focused review. Record why a disclosure is required, not required, or covered by an exception.
4. Preserve machine-readable provenance
For provider workflows, test whether generated outputs carry durable, machine-readable signals and remain detectable after common transformations such as resizing, transcoding, screenshots, or platform upload. The Commission says solutions should be effective, interoperable, robust, and reliable as far as technically feasible.
5. Add clear, accessible visible labels
Place disclosures where people will encounter them before being misled—not only in a distant policy page. Use plain language and test visibility on mobile, desktop, embeds, and reposts. The EU has published a set of AI-content labeling icons that deployers may use.
6. Define human-review and editorial-responsibility records
If relying on a human-review or editorial-responsibility exception for public-interest text, record who reviewed it, what was checked, when approval occurred, and who accepted editorial responsibility. A checkbox with no review evidence is weak operational proof.
7. Update vendor and creator instructions
Require agencies, creators, freelancers, and software vendors to disclose AI use, preserve provenance metadata where available, and follow your labeling rules. Define how they must report mistakes and supply evidence.
8. Run a publication test
Create representative test assets for text, image, audio, and video. Verify the machine-readable signal, visible label, accessibility, language, placement, and behavior after export and platform upload. Save screenshots and test results.
9. Create a monitoring and correction process
Track unlabeled content, removed metadata, false positives, user complaints, and platform changes. Establish a correction workflow and review the Commission’s guidance as technical standards evolve.
Copy-paste audit record
AI content use case:
Business owner:
System/model/vendor:
Provider, deployer, or both:
Content type: text / image / audio / video
Public-interest content involved: yes / no / uncertain
Deepfake or realistic synthetic media involved: yes / no / uncertain
Machine-readable marking present:
Visible disclosure required:
Disclosure wording and placement:
Human reviewer and review date:
Editorially responsible person:
Transformation/upload tests completed:
Evidence location:
Exception relied upon, if any:
Legal/compliance approval:
Next review date:Official resources
- Meta: signing the EU AI Act transparency code
- European Commission: Code of Practice on Transparency of AI-Generated Content
- Full Code of Practice (PDF)
- EU AI Act Article 50
- Commission guidelines on transparency of AI-generated content
- How providers and deployers can sign the code
FAQ
Is the EU AI-generated content code mandatory?
The European Commission says adherence to the code is voluntary, but the relevant Article 50 transparency requirements are legal obligations. The code offers a recognized framework for signatories to demonstrate compliance.
When do the Article 50 transparency obligations apply?
The Commission’s official page states that they apply from August 2, 2026.
Does every AI-assisted article require a label?
Do not assume that every use of AI has the same disclosure requirement. Article 50 covers specific providers, deployers, content types, and circumstances. Public-interest text, deepfakes, machine-readable marking, human review, and editorial responsibility require separate analysis.
What does Meta’s announcement change for Facebook and Instagram users?
Meta’s July 28 announcement confirms its intention to sign the code and continue its AI-content identification and labeling work. It does not by itself specify a new action that every Facebook or Instagram user must take. Businesses should follow current platform disclosure tools and their own legal assessment.
Last checked against the cited official sources on July 28, 2026.