Skip to content
DraconiticResearch
Search Sources API
203,948 studies31 sourcesUpdated 58 min ago
PreprintCC BY 4.0Zenodo

The Accountability Axis for Agent Actions: A Verifiable Form (v0.1.0)

  • Lunina K.
Open at sourcePDFDOICite

Abstract

This document specifies the accountability axis (α) for AI agent actions as a verifiable form. It introduces α as a delegability predicate separate from the consequence axis (κ), defining two classes: delegable and human_only. The specification establishes a trust boundary requiring a trusted declaration outside agent control, enforces a fail-closed resolution (undeclared defaults to human_only), and distinguishes an approval from party status. It mandates a human-owned binding step for human_only actions and requires external plan-stage chain establishment with cumulative re-establishment for extensions. Verification yields three distinct outcomes: PASS, FAIL, or UNESTABLISHED. The specification is accompanied by 95 conformance test vectors and an immutable GitHub snapshot.

Related

  1. Proof carrying agent actions enable verifiable licensing of language model tool use
    Nurgali Kadyrbek, Madina Mansurova
    Scientific ReportsPeer-reviewedCC BY 4.0Full textCrossref
  2. No Action Without Proof: A Control Plane with Verifiable Constraints, Trust Aware Coordination, and Tamper Evident Accountability
    Zohaib Qamar
    ZenodoPreprintCC BY 4.0Full textZenodo
  3. PigGPA: A Domain-Constrained Agent for Porcine Genomic Breeding with Verifiable Tool Execution
    C. Liang, Y. Zhang, W. Wei +2
    bioRxivPreprintCC BY-NC-NDFull textBioinformatics & computational biologybioRxiv
  4. A semi-autonomous multi-agent pipeline for high-energy astrophysics research with enforced provenance and verifiable self-extension
    Fang Zhou
    arXivPreprintCC BY 4.0Full textHigh-energy astrophysicsInstrumentation & methodsarXiv
  5. Governance-aware autonomous retail coordination in artificial intelligence cities using multi-agent reinforcement learning, blockchain accountability, and federated learning.
    Toqeer Ali Syed, Ali Akarma, Shahid Kamal +3
    Frontiers in artificial intelligencePeer-reviewedCC BYFull textEurope PMC

Cite

BibTeXRIS
BibTeX
@misc{lunina2026accountability,
  title = {{The Accountability Axis for Agent Actions: A Verifiable Form (v0.1.0)}},
  author = {Lunina K.},
  howpublished = {Zenodo},
  year = {2026},
  month = oct,
  doi = {10.5281/zenodo.23462036},
  url = {https://zenodo.org/records/23462036},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - The Accountability Axis for Agent Actions: A Verifiable Form (v0.1.0)
AU  - Lunina K.
PY  - 2026
DA  - 2026/10/19
PB  - Zenodo
DO  - 10.5281/zenodo.23462036
UR  - https://zenodo.org/records/23462036
AB  - This document specifies the accountability axis (α) for AI agent actions as a verifiable form. It introduces α as a delegability predicate separate from the consequence axis (κ), defining two classes: delegable and human_only. The specification establishes a trust boundary requiring a trusted declaration outside agent control, enforces a fail-closed resolution (undeclared defaults to human_only), and distinguishes an approval from party status. It mandates a human-owned binding step for human_only actions and requires external plan-stage chain establishment with cumulative re-establishment for extensions. Verification yields three distinct outcomes: PASS, FAIL, or UNESTABLISHED. The specification is accompanied by 95 conformance test vectors and an immutable GitHub snapshot.
ER  -