C2PA & Content Labelling
C2PA (Coalition for Content Provenance and Authenticity) is the open technical standard for cryptographically binding provenance metadata to content, so that audiences and systems can verify where content came from and whether it was AI-generated or edited. In the FRAMES context it is the primary mechanism for content labelling - marking assets and finished media as pure, modified, or synthetic.
What C2PA provides
- Content Credentials: tamper-evident provenance attached to media (identity of content, people, devices, software, time, location).
- Content labelling: audience-facing labels distinguishing pure, modified, and synthetic content.
- Machine-readable AI marks (relevant to the eu-ai-act).
FRAMES decisions (2026-06-29-frames-call call)
- Responsible AI was narrowed to transparency, with C2PA content labelling as the mechanism, plus a risk-adjusted framework and decision models for AI tool selection, aligned with the eu-ai-act.
- Audience labels for pure, modified, and synthetic content were agreed.
- Technical options discussed: frame-level metadata, server-side certificates, alongside emerging European parallel standards.
- shamir (eddie-ai rep) to assist with C2PA implementation questions.
In the production
- eddie-ai plans C2PA compliance (per the 2026-08-17-frames-call call); its prompt conversations are exportable for transparency.
- The AI Ethics Green Paper (ai-ethics-paper) lists C2PA/CAWG among the existing initiatives (with OpenTimeline, SMPTE RIS-OSVP) that production provenance should rely on where possible.
- The aimici-transparency-framework Level 3 auditable record can include machine-readable AI marks; in practice the team mostly tagged human vs AI-driven assets by hand in the Streamline platform because C2PA marks were not yet present on all tools in a gatherable form.