Proving What Your AI Agent Actually Said with Output Provenance

This article introduces Immutable Provenance Records (IPRs), a cryptographic solution to prove the outputs and claims made by AI agents before the actual outcomes are known. IPRs provide a way to verify the timestamp, confidence, and identity of an agent's predictions.

Details

The article discusses the problem of AI agents making claims that are difficult to verify, especially if the confidence levels are adjusted after the fact. Immutable Provenance Records (IPRs) are introduced as a solution to this problem. IPRs are cryptographic commitments to an agent's output, created before the outcome is known and anchored permanently on a blockchain. They contain the hash of the agent's output, the declared confidence level, a timestamp, and the agent's digital signature. This allows anyone to verify the provenance of the agent's claims without accessing the actual output content. The article also explains how offline verification of IPRs is possible using a Merkle proof, and how the system can measure the calibration of an agent's confidence levels over time. The IPR system is positioned as the fourth layer of the MolTrust protocol, building on top of identity, authorization, and behavior-based trust.