Execution-Attested Side-Effect Gating at the NLIP Type-3 Boundary
Open-source AI agents increasingly invoke side-effecting tools such as code publication and repository mutation across standardized communication layers like the Natural Language Interaction Protocol (NLIP, ECMA-430), where agent self-reports and post-check working-tree edits can bypass prompt-level instructions. This report presents nlip-receipts, a deterministic Type-3 Security Extension for NLIP that permits publication if and only if an authorized caller references an immutable Git tree digest previously verified inside an isolated archive snapshot under the active policy epoch and destination profile, and maps the mechanism to ECMA-434 security countermeasures.
Authors
- Kenny Wang
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24297976
- Primary Topic
- Security and Verification in Computing
- Type
- preprint