Verifiable evidence
A standard for signed evidence you can verify yourself.
Typed Standards defines a format for publishing AI-generated answers as evidence packages that carry their own cryptographic proof — a signature, a content hash, and a public trust registry. Anyone can check a package independently, without trusting the site that produced it.
Signed at the source
Every package is signed and hashed when it is produced. The signature covers the exact bytes, so any later change is detectable.
Verify independently
The verifier resolves a package's own proofs and re-checks them in your browser — it doesn't take the publisher's word for it.
Shows the math
Each check renders as it resolves: the recomputed hash, the signature result, the registry lookup. Integrity and identity — not a claim of correctness.
In plain language
Typed Standards is an open standard for independent, third-party verification of AI-generated answers. It gives someone else a cryptographically signed way to confirm that an answer is reproducible and verifiable from its inputs — without attesting that any specific answer is correct. Standardized provenance and reproducibility of AI output give others a baseline to evaluate trustworthiness independently, on their own criteria.
Civic AI Tools is the civic reference implementation of Typed Standards. See the roadmap for where the standard stands.
Verification surfaces integrity, identity, timestamp, and transparency — whether a package is intact and who signed it. It does not judge whether the content is correct. The full specification is in pre-launch review and is not published here yet.
Working on something this could serve, or want to follow along? Get in touch.