GPAI Code of Practice vs the AI-content transparency code
The Commission FAQ says Article 50 complements GPAI rules: the GPAI code covers model documentation, the transparency code covers marking and labelling outputs.
For a generated video, the transparency code is the one about the output. The Commission's Transparency obligations under Article 50 of the AI Act page says the GPAI Code of Practice "focuses on GPAI models and documentation and information" for the AI Office, national authorities and downstream providers, plus training data transparency, while the Code of Practice on transparency of AI-generated content "addresses marking techniques and transparency of the AI generated or manipulated outputs towards the persons exposed to them." This reports the page and is not legal advice.
How do the two codes relate?
The FAQ says the Article 50 transparency obligations are complementary to the rules for general-purpose AI models in Articles 53 and 55. The means to comply with those are detailed in the GPAI Code of Practice and the Commission's template for the summary of training content.
The transparency code works at system level, including but not limited to general-purpose AI systems. The page adds that transparency techniques GPAI model providers can implement to help downstream system providers meet Article 50 will also be considered in the context of the transparency code.
| Question | GPAI Code of Practice | Transparency of AI-generated content code |
|---|---|---|
| Subject | General-purpose AI models | Outputs of generative AI systems |
| Articles | 53 and 55 | 50(2), (4), (5) |
| Audience for the information | AI Office, authorities, downstream providers | People exposed to the output |
| Typical content | Model documentation, training data summary | Marking, detection, labelling of deepfakes and text |
Which one applies to my video workflow?
If you make and publish video, the marking and labelling code is the closer fit, and Section 2 is the deployer side, per the Code of Practice on Transparency of AI-generated Content page. The GPAI code matters more to whoever develops the underlying model. A downstream provider may rely on what the model provider supplies, which is why the FAQ mentions techniques that help downstream providers.
Does Sume tell me which model ran?
Not always. Sume's Videos docs say sume/auto lets Sume pick the model family, responses echo sume/auto, and the family that ran is never disclosed. If your records need the model, name one explicitly rather than using sume/auto, and keep the job id from Jobs and results.
Sources
- European Commission: Transparency obligations under Article 50 of the AI Act (read 2026-10-02)
- European Commission: Code of Practice on Transparency of AI-generated Content (read 2026-10-02)
- European Commission: Guidelines on transparency obligations for providers and deployers of certain AI systems (read 2026-10-02)
Related posts
More in Use cases
- Event recap: trim dead pauses from a talk with inspect and trim
YouTube's coming editing assistant promises to trim pauses. Until then, find silent gaps in a talk from transcript segments and cut keeper ranges with Sume.
- Event recap video from phone clips: trim, join, add music
Make a 45-second recap from phone footage: trim each clip at $0.02, join with fades in one Timeline render at $0.10, and add a $0.125 music bed.
- Flow's 50 daily credits don't roll over: plan a production week
Flow refreshes 50 free credits a day after first use and drops what you skip. How to plan a week of clips, and how a Sume USD balance behaves instead.
- Flow monthly credits by Google AI plan: how many clips fit
Flow gives 200, 1,000, 10,000 or 25,000 bonus credits a month by plan. What that buys in Veo 3.1 and Gemini Omni Flash clips, and how a Sume balance differs.
Written by Sume