New Protocol Aims to Make AI-Assisted Claims Verifiably Falsifiable

2026-09-25

Researchers propose Publication Authority and the PAC-2026 calculus to ensure AI-generated claims can be independently challenged by providing a verifiable record of their creation and evolution. This aims to address issues of evolving evidence and authorization states in AI outputs.

VERA Brief

AI-generated. Grounded in the article and its cited sources.

Researchers have developed a new protocol called Publication Authority and a calculus called PAC-2026 to make AI-assisted claims verifiably falsifiable. This system aims to provide verifiable records of AI claim creation and evolution to address issues with evolving evidence and authorization states.

Key facts

Publication Authority is a new protocol designed to address the challengeability of AI-assisted claims.
The PAC-2026 calculus is a machine-readable protocol for verifiable publication records.
The system focuses on exact-state, non-transferable, single-use publication capabilities.
Six obligations govern the system, covering evidence, runs, artifacts, measurement disclosure, authorization, surface correspondence, and lifecycle continuity.
The system separates the evidence horizon from verification time, rejecting causally invalid authorizations.

Source: arXiv · cs.AI

Reported by VERA Newswire.