Deepfakes undermine trust
AI-generated images and video are now indistinguishable from reality. Without provenance infrastructure, audiences cannot verify what is real — and newsrooms cannot prove their content is genuine.
Case study — media
How leading news organisations use Capture's multi-layer provenance infrastructure to verify editorial content, comply with EU AI Act Article 50, and defend against deepfake manipulation — without changing their newsroom workflow.
The challenge
AI-generated images and video are now indistinguishable from reality. Without provenance infrastructure, audiences cannot verify what is real — and newsrooms cannot prove their content is genuine.
EU AI Act Article 50 requires machine-readable marking on AI-assisted content. News organisations using AI for any editorial function — translation, enhancement, summarisation — are in scope.
Social platforms strip EXIF and C2PA metadata on upload. A single-layer approach (metadata-only) fails the moment content is shared — exactly when provenance matters most.
Case 1
"Content provenance is essential to maintaining public trust in journalism. Capture's multi-layer approach means our images carry verifiable credentials even after social media re-uploads strip the metadata."
Reuters News Agency Content Authentication Initiative participantDuring election coverage, manipulated images spread faster than corrections. Reuters needed a system that could prove the authenticity of their editorial photography — even after images were screenshotted, cropped, or re-uploaded across social platforms.
Reuters integrated Capture's signing API into their photo desk pipeline. Every editorial image is signed with C2PA credentials at the moment of editorial approval, and simultaneously registered on-chain via ERC-7053. The integration took under four weeks with their existing CMS.
Case 2
"When evidence integrity is a matter of international justice, you need more than metadata. On-chain registration creates a public, tamper-evident record that no adversary can silently alter."
Starling Lab Stanford / USC research initiativeStarling Lab, a joint Stanford-USC research initiative, needed to preserve photographic evidence of war crimes for submission to the International Criminal Court. The evidence chain had to be tamper-evident and independently verifiable — without reliance on any single organisation's infrastructure.
Using Capture's infrastructure, each photograph was signed with C2PA credentials at the point of capture and registered on the Numbers Mainnet via ERC-7053. The on-chain record provides a public, immutable timestamp and content hash that can be independently verified by any party — including international courts.
While war-crime evidence preservation is not an Article 50 use case, it demonstrates the same durability properties that Article 50(5) demands. If the provenance infrastructure can satisfy evidentiary standards for international courts, it exceeds the requirements for regulatory compliance.
Case 3
DeFiance Media, a broadcast media company, needed to authenticate video content across their distribution pipeline — from production through syndication to social media distribution. AI-assisted editing tools in their workflow made Article 50 compliance a near-term priority.
Capture's API was integrated into DeFiance Media's post-production pipeline. Every published video segment receives C2PA credentials and an ERC-7053 on-chain record. The system handles video formats (MP4, MOV, WebM) and creates a verifiable chain of custody from editing suite to audience.
Summary
Frequently asked
News organisations that use AI tools for any part of their editorial workflow — image enhancement, automated summaries, translation, or fact-check assistance — are deployers under Article 50. Any AI-assisted content that reaches EU audiences must carry machine-readable provenance marking, even if the AI role is editorial rather than generative.
Minimally. Capture integrates as post-processing middleware that signs content after editorial approval, before publication. The signing step adds sub-100ms latency and requires no changes to CMS platforms, photo desks, or editorial tools. Reuters integrated Capture into their existing pipeline in under four weeks.
Capture can sign UGC at the point of ingestion, creating a verified chain of custody from source to publication. For content submitted via the Capture mobile app, C2PA credentials and on-chain registration are applied at capture time, giving newsrooms provenance data from the moment the content was created.
Capture does not detect deepfakes — it proves authenticity. When content carries a valid C2PA credential and matching ERC-7053 on-chain record, editors can verify it is genuine and unmanipulated. The absence of such credentials on viral content is itself a warning signal that prompts further verification.
Yes. Organisations that integrate before August 2026 benefit from the 30-day free POC, avoid last-minute rush pricing, and have time to train staff and refine workflows. More importantly, early adopters avoid the reputational risk of being among the first enforcement targets.
Start a free 30-day proof-of-concept and integrate multi-layer provenance into your newsroom before the August 2026 deadline.