What is C2PA? Content Credentials, explained — and what they can't do for writing

The short answer

C2PA — the Coalition for Content Provenance and Authenticity — is an open technical standard that lets cameras, editing software, and AI generators attach a cryptographically signed history to a piece of media: who or what made it, and what was done to it since. The consumer-facing name for that history is Content Credentials, and the coalition describes them as functioning "like a nutrition label for digital content."

It is the closest thing the industry has to a shared answer to "where did this image come from?" — and it matters here because writing provenance, the problem Authmarks works on, is the same question applied to a medium C2PA was not designed to watch. Understanding what C2PA records, and what it openly says it cannot record, is the fastest way to understand what a provenance record for an essay has to do differently.

Who is behind it

C2PA was founded on February 22, 2021 by Adobe, Microsoft, Truepic, Arm, Intel, and the BBC, as a Joint Development Foundation project under the Linux Foundation. Its own description: "an open technical standard for publishers, creators and consumers to establish the origin and edits of digital content."

The steering committee today reads like a list of everyone with a stake in the synthetic-media problem: Adobe, Amazon, BBC, Google, Meta, Microsoft, OpenAI, Publicis Groupe, Sony, TikTok, and Truepic. Google announced its steering-committee membership in 2024; a September 2024 Google blog post says "earlier this year, we joined the Coalition for Content Provenance and Authenticity (C2PA) as a steering committee member."

The standard is also on a path to formal international standardization: the specification entered the ISO fast-track process as ISO/DIS 22144 (in process as of this writing).

How it works, in one paragraph

A C2PA-enabled tool — a camera, Photoshop, an AI image generator — writes a manifest into the file. The manifest holds a set of assertions (what device captured this, what edits were applied, whether generative AI was used), and those assertions are, in the spec's words, "digitally signed to ensure their authenticity and integrity." Each subsequent C2PA-aware edit appends to the chain. Anyone with a validator can then read the history and check the signatures. That is the whole idea: not a detector guessing after the fact, but a signed record written at creation time — the same architectural bet Authmarks makes for documents, which we unpack in what is writing provenance.

Where it already ships

This is not a paper standard. In October 2023, the Content Authenticity Initiative announced the Leica M11-P as the "world's first camera with Content Credentials built-in" — provenance signed in hardware at the moment of capture. OpenAI added C2PA metadata to DALL-E 3 images (with the important caveat, noted at the time, that the metadata is easily removed — more on that below). And in May 2024 TikTok said it was the "first video sharing platform to implement" Content Credentials, both labeling incoming AI-generated content and attaching credentials to content made in the app.

What C2PA says it cannot do

The spec's explainer is unusually honest, and three of its caveats matter for anyone tempted to treat provenance as a truth machine.

It does not judge truth. "Content Credentials do not provide value judgments about whether a given set of provenance data is 'true.'" And: "Provenance information alone cannot tell you whether the digital content is true, accurate or factual." A signed record proves the history was recorded and not altered — not that the content is honest. We make the identical disclaimer about writing records in what a writing record cannot prove.

The metadata can be stripped. Asked whether provenance data can survive removal, the explainer's answer begins "Yes it can" — but the mechanism is telling: "durable Content Credentials" combine "hard binding (aka cryptographic hashing) with a soft binding (e.g., watermarking and fingerprinting)" so a stripped file can be re-matched to its manifest. In other words, a plain C2PA manifest is removable — screenshot an image and the credentials are gone — and the standard's answer is recovery, not prevention. OpenAI's DALL-E metadata drew exactly this criticism.

Absence proves nothing. Because most content has no credentials and credentials can be removed, a file without Content Credentials is not suspicious. Any system built on signed provenance — C2PA's or ours — can only ever prove the positive case.

Why writing needed a different design

C2PA answers "what device or tool produced this file, and what edits touched it?" For an essay, the question an integrity dispute actually turns on is finer-grained: which characters did the human type, which were pasted, which were machine-suggested — and when? A Word file's "editing history" in the C2PA sense would tell you Word touched it, not whether the paragraphs arrived by keystroke or by paste.

Writing provenance therefore has to watch the process, not stamp the file: a consented session record that labels typed, pasted, and AI-assisted text as each arrives, sealed so it cannot be edited afterward. That is what Authmarks builds, and where the two worlds meet is the seal itself — Authmarks embeds a C2PA-style signed manifest when sealing a document to PDF, so the final artifact carries its provenance the same way a Leica photograph does. What a verifier can and cannot conclude from that seal is the subject of what checking a seal actually shows, and anyone can run that check at /verify without an account.

The industry consensus C2PA represents — Adobe, Google, Microsoft, OpenAI, and TikTok agreeing that recorded-at-creation provenance beats after-the-fact detection — is the same consensus slowly arriving in academic integrity, one false accusation at a time. The media world got its nutrition label first. Writing is next.