Article 50 control matrix

Article 50 and Code of Practice control matrix.

Operational mapping of Article 50 transparency scenarios and the 10 June 2026 Code of Practice to SUNURA controls, evidence outputs, and customer responsibilities.

Legal boundary

The Code of Practice is voluntary and subject to Commission and AI Board adequacy assessment. SUNURA supports operational evidence for Article 50(2), 50(4), and 50(5). It does not provide legal advice or certify legal compliance.

AreaRequirementTriggerSUNURA controlEvidence generatedCustomer responsibilityStatus
Article 50(1)AI interaction disclosureA person interacts with an AI system such as a chatbot or AI assistant.Disclosure policy, persistent notice, runtime API, acknowledgement, and gate evidence.Notice hash, render event, acknowledgement evidence, and gate event.Embed and operate the notice in the real product.Supported operational workflow
Article 50(2) / Code Section 1Provider-side machine-readable markingA provider generates or manipulates text, image, audio, or video output.Watermark/provenance workflow, C2PA-compatible evidence model, machine-readable metadata, and marking evidence route.Watermark ID, provenance record, C2PA status, detection or verification evidence, and state-of-the-art review log.Configure production marking, signing, and detection stack where claimed.Supported where configured
Article 50(2) / Code Section 1Detection reliability and robustnessGenerated media should remain detectable as far as technically feasible.Detector-assurance flags, verification APIs, watermark verification, and robustness evidence records.Detector method, limitations, tamper test, and versioned evidence.Select and validate trusted detector or marking providers for production claims.Readiness support only
Article 50(4) / Code Section 2Deepfake labellingAI-generated or manipulated image, audio, or video could falsely appear authentic.Deepfake review workflow, publication workflow, EU icon or equivalent label profile, and public verification page.Classification, label profile, first-exposure evidence, accessibility evidence, and publication review.Human review and lawful publication decision.Supported operational workflow
Article 50(4) / Code Section 2Public-interest AI textAI-generated or manipulated text informs the public on matters of public interest.Publication workflow, public-interest classifier, and editorial-control exception record.Editorial owner, reviewer rationale, notice snippet, or exception evidence.Record editorial responsibility and publish the notice where required.Supported operational workflow
Article 50(4) / Code Section 2Creative, satirical, or fictional deepfake contextCreative or satirical synthetic media is published.Limited-disclosure workflow that records non-hampering placement.Creative-context rationale, limited notice, and review record.Make a lawful contextual decision and avoid hiding the disclosure entirely.Evidence organisation only
Article 50(5)Clear, distinguishable, and accessible noticesNatural persons first see AI-generated or manipulated content.Runtime evidence capture, DOM or screenshot evidence, ARIA or alt fields, and label-placement proof.First-exposure evidence, accessibility evidence, and visibility evidence.Check the actually implemented surface.Supported where configured
Code signatory readinessProvider/deployer section selectionAn organisation may sign Section 1, Section 2, or both depending on its role.Role classifier, readiness mapping, and signatory evidence pack.Role decision, section mapping, and unresolved-gap register.Senior authorised signatory and legal review.Readiness support only
Adequate other meansNon-signatory evidence positionA non-signatory may need to demonstrate adequate measures if asked.Code gap register, evidence exports, control matrix, and public verification.Evidence pack, control map, and audit trail.Maintain evidence and respond to authority or customer requests.Evidence organisation only