Autonomous software agents can act through tools, delegate work, and replace their models or execution environments while earlier authorizations remain in flight. To address this, we introduce Assurance-Conditioned Continuous Authorization (ACCA), a grant‑independent continuous‑authorization approach that maintains operational authority based on currently justified assurance. ACCA separates evidence validity from contextual applicability, derives assurance claims from eligible evidence, and solves a multidimensional effective‑authority envelope bounded by a principal‑declared ceiling and applicable policy. A typed dependency graph enables selective re‑evaluation only when authorization‑relevant changes occur, ensuring efficient updates. An authority epoch advances whenever effective authority changes; an optional authorization‑basis version supports revalidation of action‑specific conditions. Protected effect‑time enforcement can reject stale actions before a consequential resource change, provided a commit mechanism exists. A compositional Domain Governance Profile expresses domain‑specific semantics without granting authority. Controlled scenario tests implemented as a regression suite using a trading profile exercise default‑deny rules, evidence‑conditioned contraction and recovery, policy divergence, selective non‑response, and stale‑action rejection at a local gateway before the simulated‑broker boundary. The study provides implementation‑level evidence in one domain, not a cross‑domain guarantee of harmful‑behavior prevention.
Review: ACCA’s clear separation of evidence and context, combined with a typed dependency graph, offers a practical and auditable way to keep autonomous AI agents continuously authorized while adapting to dynamic environments.