A visible overlay reading “AI-generated” informs a viewer directly.
Three ideas that must not be conflated
A platform watermark may be embedded into media by a specific vendor and detected by specialized tools. C2PA Content Credentials are something else: the technical specification defines signed assertions and content bindings used to validate provenance information associated with a media asset. An editor needs to know which evidence is being preserved. Removing a visible text overlay for editorial layout does not imply the cryptographic provenance survives a transcode, and a watermark does not grant permission to use the depicted person’s face.
What a valid credential can establish
The C2PA 2.2 specification describes a manifest containing signed claims about the asset, the claim generator and referenced ingredients. A verifier can check bindings and signatures under an applicable trust framework. A valid credential can support statements about the recorded creation or editing history. It does not prove that an alleged historical event took place, that a speaker’s words are truthful, or that all upstream content was licensed. Google’s guidance explicitly distinguishes a valid asset history from factual reality.
Why metadata disappears in publication pipelines
Editing, recompression, screenshots and social platforms can alter or strip the file or its embedded credentials. The C2PA specification also allows external manifests where an embedded manifest is impractical. Therefore test the exact delivery route: generate or edit, export a master, pass it through the intended transcoder, upload to the target platform, and inspect the resulting public asset if the platform exposes one. When credentials fail to carry through, describe the missing chain rather than assuming the output is deceptive or authentic.
A practical provenance record
Keep the original upload, all licensed ingredients, generation settings, editor version, consent approvals, exports and signed manifests if available. At every step compute an ordinary content hash as an internal integrity check; such a hash is not itself a C2PA credential. Link final editorial approval to the exact distributed file. Do not make a single mutable cloud project the only source of evidence. Access controls should protect both the media and any sensitive identity details in the manifest.
Disclosure for recreated news footage
A newsroom illustrating a past financial event can publish an explicitly labeled synthetic reconstruction without pretending to have actual archival footage. Its disclosure should tell viewers what is simulated, not merely add a generic vendor logo. Preserve the prompts and editorial fact-check independently from the C2PA provenance record. The existence of a valid manifest that identifies a generative tool is consistent with the scene being fictional. A verified creation chain and a checked historical claim are complementary but different quality checks.
Boundaries on alteration and removal
This guide covers preserving and validating provenance, not evading forensic marks or bypassing rights notices. Do not remove protective disclosures to mislead an audience, and do not claim credentials are guaranteed to remain intact across platforms. When a received clip has no credential, avoid inferring who created it from absence alone. Investigate source publication, chain of custody and corroborating documentation before using it as evidence.
Production choice and further reading
If a platform demands a prominent on-screen disclosure, add it according to that platform’s current rules, even if C2PA metadata is also present. If a documentary client requires cryptographic provenance, choose a compatible signing and validation workflow and test its export path. C2PA 2.2 is the technical basis used here; jurisdiction-specific disclosure law is outside this team’s scope and belongs with the policy editors.
Sources & Further Reading
- C2PA — Content Credentials Technical Specification 2.2 (first-party source; checked 2026-10-25)
- Google — Understand AI labels and Content Credentials limitations (first-party source; checked 2026-10-25)
Draft editorial material. No independent hands-on test, fabricated accuracy benchmark, or third-party media license is claimed. Product eligibility and commercial terms should be rechecked immediately before publication.