Guardat en:
| Autor principal: | |
|---|---|
| Format: | Recurso digital |
| Idioma: | |
| Publicat: |
Zenodo
2026
|
| Matèries: | |
| Accés en línia: | https://doi.org/10.5281/zenodo.20331819 |
| Etiquetes: |
Afegir etiqueta
Sense etiquetes, Sigues el primer a etiquetar aquest registre!
|
Taula de continguts:
- <div> <p class="MsoNormal"><span>Current AI governance frameworks evaluate compliance, auditability, and model behavior in isolation from operational conditions. Operational AI systems, however, increasingly fail through structural degradation that remains invisible to traditional governance instruments — they drift, delegate incorrectly, and continue executing after governance capability has effectively collapsed.<span> </span>This paper introduces PCS-01 (Provider Compliance Simulator), a runtime governance stress environment designed to simulate and observe governance survivability under operational conditions. Rather than evaluating whether AI outputs are correct, PCS-01 evaluates whether governance remains executable during runtime.<span> </span>The architecture models five primary failure modes: Drift Accumulation (FM-01), Escalation Failure (FM-02), Carrier Loss (FM-03), Recursive Delegation (FM-04), and Silent Corruption (FM-05). PCS-01 introduces the concept of Governance Survivability as a measurable runtime property, distinct from auditability and policy compliance, and operationalises it through a four-gate Admissibility Engine (Q1–Q4), a Human Commit Boundary (HCB), and a three-level recovery architecture (R1–R3).<span> </span>The system is positioned as a runtime diagnostic architecture for operational AI systems in telecommunications, healthcare, critical infrastructure, and other high-consequence regulated environments operating under the EU AI Act (2024) and comparable governance frameworks. A companion software demonstrator (PCS-01 Runtime MVP) is published separately on Zenodo.</span></p> </div>