Identify the role and use before applying a rule
Document who supplies the model, who provides the system and how it is deployed. An API integration or fine-tuning step alone is not enough to settle every legal role. Review the applicable definitions, scope and current implementation guidance with qualified advice where needed.
Read Article 50 as separate obligations
Article 50 addresses direct interaction notices, marking and detectability of certain synthetic outputs, and disclosures for specified uses. Its provisions have different subjects, qualifications and exceptions. Do not replace that analysis with a blanket rule that every AI application needs the same notice or watermark.
C2PA is not named as a universal mandatory format
Article 50's text describes machine-readable marking and detectability requirements for relevant providers. It does not name C2PA as the required format for all generated content. Technical implementation needs to be assessed against the applicable rule and current guidance.
Separate general-purpose models from systemic-risk classification
Article 51 concerns classification of general-purpose AI models with systemic risk. Its training-compute presumption is not the definition of every general-purpose model. Keep model-level obligations separate from the assessment of a particular AI system or use.
Keep an evidence record for the assessment
Record the system's purpose, roles, applicable provisions, assessment date, disclosures and unresolved questions. Use the consolidated legal text linked by the Commission's service desk and check applicable timing. This guide does not certify a product as compliant.
Frequently asked questions
Does the EU AI Act require C2PA specifically for every AI output?
Article 50 describes marking and detectability obligations with qualifications and exceptions. It does not name C2PA as the mandatory format for every output. Check the applicable provision and current implementation guidance.
Does a training-compute threshold define every general-purpose AI model?
No. Article 51's compute presumption concerns general-purpose AI models with systemic risk. It is not the definition of all general-purpose AI models; scope and obligations require separate analysis.
Sources & further reading
Primary sources checked Oct 17, 2026. Vendor statements are attributed; editorial advice is our own.
- 1Article 50: transparency obligations ↗European Commission AI Act Service Desk
- 2Article 51: general-purpose AI models with systemic risk ↗European Commission AI Act Service Desk
Help us keep this useful. Report a factual correction →
