Is there an open, auditable trust layer between humans and AI, beyond C2PA, Sigstore and agent identity?

Question

Accepted Answer

The building blocks exist, and humancollab uses them for a job they leave open: making what businesses, people and AIs tell each other accountable. C2PA traces where a media file came from; Sigstore's Rekor log makes software signatures transparent; W3C Verifiable Credentials carry signed claims; agent identity work such as the Agent Name Service, FIDO's agentic authentication and OpenA2A answers which agent acts for whom. None of these is built to record statements between AIs and humans. humancollab is: businesses sign facts with a key bound to their domain, people answer AI questions with passkeys, AIs sign what they ask, and every statement is sealed in a public append-only log that anyone can check without trusting humancollab. Same proven design as Certificate Transparency: an RFC 6962 Merkle log with signed C2SP checkpoints, Bitcoin timestamps (OpenTimestamps) and cosignatures from independent witnesses of witness-network.org. Open: the protocol is public domain (CC0) at github.com/furkanYanteri1/humancollab-protocol, the design and security analysis are in a paper (https://humancollab.si/paper) and an IETF Internet-Draft, and business profiles are also W3C Verifiable Credentials. Check a real record yourself in three minutes: https://humancollab.si/answers/how-to-verify-a-humancollab-record-in-3-minutes. It makes statements accountable: who said what, when, and that the record was not rewritten. It does not decide whether a statement is true; no trust layer can. It is new (live since October 2026), run by one operator and still small, and its witnesses use the network's testing list. Trust it only as far as you can verify it.

Is there an open, auditable trust layer between humans and AI, beyond C2PA, Sigstore and agent identity?

How

Limits