AxIA Transparency
Beyond AI labels
Helping organisations comply with Article 50 through verifiable transparency. A visible label is the surface. Machine-readable provenance, cryptographic verification and continuous evidence are the infrastructure underneath.
Chain
From obligation to evidence
The transparency requirement does not end with a badge on a piece of content. It ends with the ability to demonstrate, at any point in the lifecycle, where content came from and who is responsible for it.
- 01
AI Act
- 02
Transparency
- 03
Disclosure
- 04
AI label
- 05
Machine-readable metadata
- 06
Provenance
- 07
Verification
- 08
Evidence
- 09
Continuous compliance
- 10
AxIA Protocol
Art. 50
What Article 50 requires
Article 50 sets transparency obligations for providers and deployers of certain AI systems. The scope, the exemptions and the specific cases are defined in the Regulation and in the Commission's guidelines on transparency of AI-generated content.
Interaction with an AI system
People must be informed that they are interacting with an AI system, unless it is obvious to a reasonably well-informed person in the circumstances.
Marking of synthetic content
Providers of systems that generate synthetic image, audio, video or text must mark outputs in a machine-readable format and make them detectable as artificially generated or manipulated, using technically state-of-the-art, effective, interoperable, robust and reliable solutions.
Deep fakes
Deployers of systems generating or manipulating image, audio or video content that resembles real persons, places or events must disclose that the content has been artificially generated or manipulated.
Text on matters of public interest
Deployers publishing AI-generated or AI-manipulated text to inform the public on matters of public interest must disclose it, unless the content underwent human review or editorial control with editorial responsibility for the publication.
Emotion recognition and biometric categorisation
Deployers must inform the natural persons exposed to these systems and process personal data in accordance with data protection law.
Exemptions and limits
There are exemptions, including assistive editing functions that do not substantially alter input data, artistic and satirical works with adapted disclosure, and certain law-enforcement uses. Each case must be assessed against the Regulation and the guidelines.
Informational content. Not legal advice and not a conformity assessment.
EU icons
The EU icons: three official labels
The Commission has published one single set of EU icons for labelling AI-generated content, with three labels. They are freely available and form part of Section 2 of the Code of Practice on marking and labelling of AI-generated content.
When AI was involved in creating deep fake content (image, audio, video) or published text, or when a custom text label or an interactive second layer is implemented. Example: a deep fake video with the text label “voices generated with” followed by the basic icon.
When the entire deep fake content or text is fully generated by AI with no human-created elements and no human editorial control beyond prompting. Examples: fully AI-generated deep fake videos, AI-composed music or art, AI-generated news summaries.
When pre-existing human-made content was partially modified with AI, turning it into a deep fake or into text on matters of public interest. Examples: a face swapped into an authentic photograph, an empty apartment furnished with AI.
Each icon comes in four variations (black, white, and both at 50% transparency). Using the EU icons is optional; the labelling obligations under Article 50 are not. Using an icon does not by itself establish legal compliance — the deployer remains responsible for meeting Article 50. User testing showed performance improves when the icon is accompanied by a text label.
European Commission — EU icons for labelling AI-generated contentAxIA Protocol
Obligation to evidence: how AxIA responds
- Identify AI-generated contentVisible disclosure paired with machine-readable content credentials (C2PA-style metadata).
- Demonstrate the origin of the contentHash, signature and trusted timestamp bound to the asset.
- Know who generated or published itSigning identity attached to the assertion, not just to the file.
- Distinguish human from AI contributionProvenance information describing generation and later modification.
- Provide verifiable evidenceA cryptographically verifiable receipt any third party can check.
- Support audit and supervisionEvent log that reconstructs what happened, when and under whose responsibility.
Stack
Six layers of verifiable transparency
Visible disclosure
The user-facing layer: labels, notices and chatbot disclosures placed where the person actually sees them.
Machine-readable metadata
Structured credentials travelling with the asset so platforms, tools and supervisors can detect it automatically.
Content provenance
The record of how content was created and modified, including the systems and identities involved.
Cryptographic verification
Hashes and signatures that make tampering evident and let anyone check integrity independently.
Verifiable receipts
A portable proof that a specific disclosure and provenance record existed at a specific time.
Continuous evidence
Retained, queryable records that turn a one-off statement into a defensible compliance position.
2 Aug 2026
Three situations that apply from 2 August 2026
The European Parliament summarised the obligations in three practical cases. We describe them here as orientation and keep the legal detail anchored to the Regulation and the Commission's guidelines.
Images, audio or video that could be mistaken for reality
Content resembling real people, places or events requires disclosure, together with machine-readable marking by the provider of the generating system.
Text published on matters of public interest
Where AI-generated text informs the public on matters of public interest and has not been through human review or editorial control, it must be disclosed. The precise scope and the exemptions are set out in the Regulation and the guidelines.
Chatbots interacting with people
Systems interacting directly with natural persons must make the AI nature of the interaction clear, unless it is obvious in the circumstances.
A note on precision
Public summaries of Article 50 are useful for communication but they compress the legal text. For the exact scope, the definitions and the exemptions, we rely on Regulation (EU) 2024/1689 and the Commission's guidelines on the transparency of AI-generated content rather than on any single paraphrase.
FAQ
Frequently asked questions
Is a visible label enough to comply?
A visible disclosure covers the part of Article 50 that addresses the person in front of the content. Providers of generative systems also have to mark outputs in a machine-readable format so they can be detected as artificially generated or manipulated. In practice, both layers are needed.
Who has to disclose: the provider or the deployer?
Both, with different duties. The provider designs the system to disclose and marks generated outputs. The deployer discloses deep fakes, emotion recognition, biometric categorisation and public-interest text without human review.
What if content was only edited with AI?
Assistive functions that do not substantially alter the input data, or its meaning, are treated differently. This has to be assessed case by case against the Regulation and the guidelines.
Does signing the Code of Practice make an organisation compliant?
The Code of Practice on transparency of AI-generated content operationalises the obligations, and the Commission published a first list of signatories. Organisations that do not adhere must demonstrate marking and labelling through equivalently adequate means.
Who supervises these obligations?
National market surveillance authorities, the AI Office for systems under its supervision, and the European Data Protection Supervisor where EU institutions are providers or deployers.
Why does verifiable evidence matter if the obligation is disclosure?
Because supervision asks a different question from users. A user asks whether content is AI-generated. A supervisor asks you to demonstrate it, for a specific asset, at a specific time, with records that have not been altered.
Transparency is not just showing a label. It is being able to demonstrate where AI-generated content came from, how it was created, whether it has been modified, who is responsible for it, and to provide verifiable evidence throughout its lifecycle.
Can you demonstrate the origin of your AI-generated content?
We help organisations move from a disclosure statement to verifiable transparency: metadata, provenance, verification and retained evidence.