Ghost Signature: cryptographic provenance for AI outputs

AI-generated text is everywhere — but how do you prove a specific output came from your system? Ghost Signature uses Ed25519 + a public JWKS so anyone can verify authorship without trusting your server.

Ghost Signature: cryptographic provenance for AI outputs

AI-generated text is everywhere. Customer support replies, product descriptions, agent actions, internal reports — and increasingly, the content your users see was written or shaped by an LLM somewhere upstream.

That raises an awkward question: **how do you prove a specific output actually came from your system?**

Screenshots can be edited. API logs are private. "Trust me bro" doesn't scale. What you really want is a tiny piece of math that anyone — your customer, a regulator, another agent — can verify on their own.

That is what Ghost Signature does.

The shape of the problem

Most "AI watermarking" today falls into two camps:

1. **Statistical watermarks** — bias the token distribution slightly so you can detect "this came from model X". Fragile, model-specific, breaks under paraphrasing.

2. **Server-side logs** — "ask us and we'll confirm". Requires trusting one party. Useless across organizations.

Neither answers the practical question: *"Did Fyrnity's API actually produce this exact payload at this exact time?"*

What Ghost Signature does

Five endpoints, one idea: **sign the bytes you ship, publish your public key, let the world verify.**

The verifier never has to call you. They just need the JWKS URL and the signature.

Why Ed25519

If you've ever signed an `.deb` package or a JWT, you already know the shape.

The `kid` matters more than you think

Every signature carries a key identifier (`kid`). When you rotate keys — and you will — old signatures keep verifying against the old `kid` in your JWKS, while new signatures use the new one.

This is the difference between *signing* and *signing in a way that survives a key compromise three years from now*.

Soft revocation marks a `kid` as "no longer trusted for new signatures" without retroactively invalidating the legitimate history. That matters for audit trails, contract evidence, and content provenance.

Where this fits

It composes well with the rest of the stack: pair it with **ChromaWrite** to humanize text *and* prove it came from your account, or with **Agent Safety Pack** to sign killswitch state changes.

Pricing

Same shape as the rest of the Fyrnity API family:

All plans are hard-capped — you will never get a surprise bill.

Try it

Live on RapidAPI now: search **Ghost Signature** under the Fyrnity provider, or grab the key directly from the [API Tools page](https://fyrnity.com/tools).

Verifiable provenance shouldn't require a PhD or a vendor lock-in. It should be one HTTP call.

👉 **[Try Ghost Signature on RapidAPI →](https://rapidapi.com/fredrikssonanton01/api/ghostsignature)**

Loading interactive experience…