Use case
Media Authenticity and Provenance Verification
How Noesea uses cryptographic provenance and content-addressed evidence to make images, video and documents verifiable at the receiving side — provenance over deepfake detection.
On this page
The problem with detection
As generative models improve, forensic deepfake detection becomes a losing arms race: the score is probabilistic, it degrades over time, and the person receiving the media still has to trust the detector. Newsrooms, courts and platforms need something that holds up regardless of how convincing synthetic media gets.
Provenance-first verification
Noesea records assertions about where an artifact came from and how it changed. Media can be hashed, anchored under a Merkle root, and tied to custody events and C2PA-style provenance signals. A receiver can verify the bytes and any valid signatures that are actually present; origin claims, availability, and substantive factuality remain separate.
When authenticity is contested
If a media claim is challenged, the protocol settles it with stake-backed juror voting and records direction, process finality, vote counts, and dissent separately. The result is a durable, auditable record of how a contested piece of media was reviewed — not an authenticity probability or editorial fact.
What the receiver can check
Given the file and an exported proof, a receiver can recompute the content hash, confirm through a Merkle proof that the file belongs to an anchored bundle, and validate any signatures that are present. Each result is reported as its own field. A matching hash says the bytes are the ones that were recorded; it does not say the depicted event happened, who captured it, or whether the file is still stored elsewhere.
Where detection still fits
Forensic analysis is not thrown away. A detector's output can be attached as one more piece of evidence, with its model and version identified, and weighed by reviewers or jurors. It is not treated as the authority that decides authenticity, because its reliability shifts as generation techniques change.
A newsroom workflow
Hash original captures before any edit. Record each handoff — capture, transfer, edit, publication — as a custody event. Anchor the bundle root, and publish a verification link next to the story. A reader or another outlet can then check that the published file is the recorded one and see its custody path without contacting the newsroom.
Separate checks
One result per media predicate
Each field has its own status, evidence, issuer, and limitations. A passing result in one field never fills another field or establishes the depicted event.
- Byte integrity
- Establishes: Whether received bytes match a committed digest.
- Does not establish: Origin, authorship, or factual content.
- C2PA validity
- Establishes: Whether a selected manifest and signature validate under named trust roots.
- Does not establish: Truth, complete edit history, or signer authority.
- Capture
- Establishes: What a named capture method observed at a bounded time.
- Does not establish: Earlier origin or correctness of the served content.
- Transformation
- Establishes: Declared or evidenced edits linked to exact artifacts.
- Does not establish: That undeclared edits did not occur.
- Generator evidence
- Establishes: An attributed generator assertion or verified generator receipt.
- Does not establish: Authorship, consent, or substantive truth.
- Actor identity
- Establishes: The strength and scope of an actor binding.
- Does not establish: That the actor created, approved, or depicted the content.
- Time
- Establishes: A separately evidenced creation, capture, or assertion time.
- Does not establish: Place, actor, or depicted-event time.
- Place
- Establishes: A separately evidenced capture or assertion place.
- Does not establish: Where the depicted event occurred.
- Depicted event
- Establishes: Only an attributed proposition and its own evidence about what the media depicts.
- Does not establish: Anything from byte, provenance, capture, actor, time, or place checks alone.
FAQ
Common questions about media authenticity
Does Noesea detect deepfakes?
Noesea prioritizes provenance over detection. It records hash-consistent bytes, custody events, and issuer assertions with their proof status. Detection signals can be attached as attributed evidence, but neither provenance nor detection alone establishes substantive authenticity or factuality.
How does a receiver verify a piece of media?
The receiver checks the content hash against the anchored evidence record, reviews the custody chain, and confirms provenance signals — using a browser check, the verification console, or a portable report. No trust in the sender is required.
Does Noesea store the media itself?
No. Noesea anchors hashes, Merkle roots, custody events and outcomes — never raw content on-chain. The media stays in content-addressed storage (such as IPFS) while the proof record references it.
Does Noesea detect deepfakes?
Noesea prioritizes provenance over detection. It records and verifies where media came from and how it changed; a forensic detector's output can be attached as evidence, but it does not decide authenticity on its own.
Continue