The landscape
What else you could use, and when you should.
Four categories of tool get pointed at this problem. Three of them are good at something Tessera does not do. Here is the honest split, including the places where the right answer is not us.
Content provenance standards · C2PA and Content Credentials
What it is. An open specification for embedding signed provenance assertions inside a media file, with a broad coalition behind it and support appearing in cameras, editing software and generation tools.
Where we stand against it, plainly. Tessera writes and reads C2PA Content
Credentials through c2pa-rs, the C2PA reference implementation: the manifest is
embedded in the asset, not beside it, and validates against the open format.
Where we differ. C2PA is a format. It tells you how to express a provenance claim; it does not decide how a claim composes into a verdict, and it does not produce the organisational record that a reverse-burden defence needs. Those are the parts we build.
Software supply-chain signing · Sigstore and SLSA
What it is. Signing, verification and transparency logging for software artefacts — provenance for what came out of a build pipeline.
Why it is a fair comparison. The cryptographic shape is genuinely close: sign an artefact, bind it to content, let a third party verify without trusting the producer. If you are asking “why not just use Sigstore”, that is a reasonable question rather than a naive one.
Where it stops. The scope is build artefacts. It is not designed for a photograph that gets cropped, re-encoded and re-published, and it does not carry the media-specific assertions the AI Act asks for — that a given asset is AI-generated, produced by a named model, and disclosed as such. Different problem, adjacent mathematics.
Detection · classifiers that score whether something is synthetic
What it is. Models that look at an image, video or audio file and estimate the probability it was generated.
When it is the right tool. When you have unsigned content and no upstream relationship — which, today, is nearly everything. Detection is the only thing that works on material that arrived with no history attached, and any honest provenance vendor should tell you that.
Why we are not doing it. Detection is an arms race against generators that improve faster than detectors, and its output is a probability. A probability is difficult to put in front of a court, and it decays every time a new model ships. Proving originals is tractable, so that is what we build — and the record of what you did about it, which is the half of the problem detection does not touch at all. What our answer actually establishes →
Audit logging and AI governance platforms
What it is. Internal logging, policy management and evidence collection — usually a system of record inside one organisation.
Where it stops. A log proves internal consistency to someone who already trusts the system holding it. The statutory position is the opposite: s.66I(4) puts the burden on the defendant, which means the record has to convince a party with no reason to trust you and no access to your systems.
We regard these vendors as a channel rather than as competitors. Six of the organisations in our market map are in this category and we have them marked as partners.
Side by side
| Tessera v0.1 | C2PA | Supply-chain signing | Detection | |
|---|---|---|---|---|
| Detects a single changed byte | Yes | Yes | Yes | No |
| Works on unsigned content | No | No | No | Yes |
| Matches a re-encoded asset to its record | Yes | Partly | No | n/a |
| Embedded in the asset | Yes | Yes | n/a | n/a |
| Verifiable without the vendor | Yes | Yes | Yes | No |
| Designed around media edit history | Yes | Yes | No | No |
| Marks AI-generated per Article 50 | Yes | Yes | No | Infers only |
| Compact offline receipt for a third party | Yes | No | Via its log | No |
| Organisational evidence record, not per-asset | Yes | No | No | No |
The last two rows are the ones nothing else in this table does, and they are the evidence half of the product alongside the provenance half.
The axis this table hides
Every column above is a way of answering is this asset what it claims to be? That is one of the two questions Tessera exists for. The other is can you show a hostile third party what controls you had in place, and when?
No detector answers that. C2PA does not attempt it — it is a format for expressing a claim about an asset, not a record about an organisation. Audit platforms attempt it and stop at the boundary of your own systems. It matters because the UK offence puts the burden of proof on the defendant, so the record has to convince someone with no reason to trust you. Why that is a record and not a score →
The short version
Use detection when content arrives with no history. Use C2PA when you need provenance embedded in the file and read by existing tools today. Use supply-chain signing for build artefacts. Use Tessera when you need both halves at once — provenance on the asset and a record that a hostile third party can check — because the obligation you are answering puts the burden of proof on you.