Abstract
AARC specifies a machine-verifiable execution contract for tool-using AI agents using externally observable runtime evidence rather than hidden chain-of-thought. It focuses on execution ordering, immutable intent anchors, tool authorization, separated approval identities, state verification, trace integrity, and change provenance.
Research objective
The work addresses a practical assurance problem in agentic systems: important claims about tool execution should be supported by evidence that can be independently checked without requiring access to private model reasoning. The contract therefore emphasizes observable execution semantics, explicit authorization boundaries, trace integrity, and verifiable state transitions.
Resources
The full manuscript PDF is mirrored on this research site for scholarly discovery. The canonical archival record is preserved on Zenodo under DOI 10.5281/zenodo.23107336. The associated specification and reference materials are maintained in the AARC GitHub repository.
Citation
Saridakis, Stamatis-Christos. “AARC: A Machine-Verifiable Audit and Reliability Contract for Tool-Using AI Agents.” Version 1.1.1, 2026. DOI: 10.5281/zenodo.23107336.