AICOS CORE: Evidence-Bound Authorization and Effect-Level Security for Autonomous AI Agents
This 20-page research preprint presents AICOS CORE, a proposed architecture for evidence-bound authorization and effect-level security in autonomous AI agents. The manuscript introduces an Authorization-to-Effect Evidence Contract (AEEC) linking delegated intent, operation parameters, policy epochs, nonces, sink identity, and observation provenance. It distinguishes authorization decisions, dispatch records, acknowledgments, and independently corroborated external effects, explicitly preserving unknown and conflicting outcomes. The work develops mathematical definitions, conditional safety obligations, counterexamples, and a prospective evaluation protocol. It documents bounded local software checks and two fourteen-case synthetic diagnostics. These controlled observations do not establish real-world effectiveness, comparative security superiority, or production readiness. Independent external-effect validation and matched third-party benchmark evaluation remain outstanding. Author: Yasin KalafatoğluAffiliation: Independent Researcher, TürkiyeORCID: https://orcid.org/0009-0009-5063-999X Version: v1.0. This manuscript uses IEEEtran formatting; no IEEE acceptance, peer-review approval, or journal publication is claimed.
Authors
- YASIN KALAFATOGLU (ORCID: https://orcid.org/0009-0009-5063-999X)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24250778
- Primary Topic
- Access Control and Trust
- Type
- article
- Field-Weighted Citation Impact
- 0.00