TRUST & COMPLIANCE
How FakeXpose Fits Into The Content Authenticity Landscape.
Provenance standards and deepfake detection solve different problems. This page explains how FakeXpose's approach compares to C2PA and related standards, where each one is strong, and where FakeXpose covers gaps that provenance metadata alone can't.
REGULATORY TAILWINDS
Built For A Fast-Moving Standards Landscape.
NAIC Model Bulletin
Guidance to insurers on governing AI systems, including third-party vendor oversight.
Proposed FRE 707
Would apply reliability standards to machine-generated evidence. Published for comment in 2025–26 and sent back for further study in June 2026; not adopted.
OCC / FFIEC guidance
Supervisory expectations for documented AI-fraud governance and vendor oversight.
ISO/DIS 22144 (Content Credentials)
The international standardization track for C2PA v2, currently in Draft International Standard status as of 2026, not yet finalized/published.
General information, not legal advice.
THE PROVENANCE STANDARD
What C2PA Does — And What It Was Designed To Do.
C2PA (the Coalition for Content Provenance and Authenticity) is an open standard, not a product, founded in 2021 by Adobe, Arm, BBC, Intel, and Microsoft. It defines "Content Credentials" — cryptographically signed metadata embedded in a file at the moment it's created or edited, recording what tool made it, what changes were made, and by whom. C2PA supports audio formats including WAV and M4A, alongside images, video, and documents, and adoption is real and growing: cameras from Sony, Leica, and Nikon; editing tools from Adobe; and a growing list of AI platforms now sign their outputs.
PROVENANCE VS. DETECTION
A Credential Only Works If One Was Ever Attached.
C2PA's own documentation acknowledges this directly: credentials answer some questions well and leave others open — and that gap is exactly where statistical detection still matters. A few common, real scenarios show why:
The honest summary: C2PA is the strongest tool available for proving where content with a credential came from. FakeXpose is built for everything else — the vast majority of audio in circulation today, which was never signed in the first place, plus the human judgment a credential alone can't provide. One caveat that matters especially here: Verify Now is in beta and its automated results should be treated as a starting point, not a conclusion — for anything in this table where the outcome actually matters, the answer is Expert in the Loop's documented finding, not an automated score by itself.
WHY THIS MATTERS FOR SURVIVABILITY
A Credential Lives Next To The File. A Watermark Lives Inside It.
C2PA's Content Credential is metadata attached to the file container — which means it's only as durable as the file itself. Re-encoding the audio, converting the format, or re-uploading it through a platform that strips metadata (common on social and messaging platforms) can remove the credential entirely, with no trace that one ever existed.
SourceGuard takes a different approach: the marker is embedded directly in the audio signal, not the file's metadata. That makes it meaningfully more resistant to the transformations that break a metadata credential outright — re-encoding, format conversion, and routine re-uploading don't automatically erase it the way they erase a C2PA manifest.
Important — this doesn't mean unbreakable. Consistent with what SourceGuard's own page already says: heavy compression, aggressive editing, or a deliberate, sophisticated removal attempt can still degrade or defeat any audio watermark, including this one. The honest claim is more durable than metadata against ordinary handling — not indestructible against a determined adversary. Overstating this on a compliance page is exactly the kind of claim a technical reader will test and a competitor will call out.
PRIVACY AND SECURITY
Your Audio Stays Your Audio.
Recordings in a claim, a case file, or an investigation are sensitive. Here's how we handle them.
- Role-based access and audit logs in the dashboard
- Encrypted in transit and at rest
- You control retention — delete files and results from the dashboard whenever you choose
- Audio is not used to train detection models without your consent
- Not shared with third parties outside your case workflow