Abstract
Non-Effective Holding of Proposed Operations, Exact-Act and Boundary-Bound Authority, Sink-Side Reconstruction of the Actual Operation, Atomic Consumption, and Fail-Closed Recovery Semantics
ABSTRACT
A computer-implemented method and system control whether an operation proposed by an artificial-intelligence agent, autonomous system or other software component may produce an external effect. The proposed operation is held in a non-effective state. Authority for the operation is bound to a descriptor of its security-relevant attributes, such as action type, resource, destination, parameters and payload digest, and to an intended effectuation boundary, a freshness value, a policy or authorisation generation and a permitted use count. At the component that controls the external consequence, for example a network egress point, storage commit, tool dispatcher or actuator controller, a second descriptor is independently derived from the operation actually presented for effectuation rather than taken from an upstream assertion. The effect is permitted only if the authorised descriptor matches the derived descriptor under current protected state, the authority is unrevoked, fresh and bound to that boundary, and single-use authority is consumed atomically with the effect; otherwise no effect occurs. Unknown or unverifiable values are treated as unauthorised, timeouts never grant authority, crash recovery distinguishes non-performed from performed-but-unrecorded effects, and any equivalent path to the same consequence must pass an equivalent check. The arrangement addresses action, destination and parameter substitution, replay, duplicate execution, cross-boundary reuse, stale authority after policy change, compromised intermediate components, compromised agents, time-of-check/time-of-use gaps and rollback, including changes introduced when an agent legitimately re-plans after authorisation. It differs from caller authentication, scope-based tokens, rich authorisation requests, sender-constrained tokens, transaction-context propagation, workload identity and remote attestation, which may supply inputs but do not by themselves compare the authorised act with the act actually presented at the final effect boundary. This disclosure is based on PCT/IB2026/055615, filed 4 June 2026 and published 16 July 2026, which claims priority from PCT/IB2026/053385, filed 7 April 2026 and published 4 June 2026. KEYWORDS FOR EXAMINER SEARCH effectuation boundary verification, pre-effect authorization check, non-effective pending operation, exact-act binding, action descriptor matching, sink-side reconstruction of operation, boundary-bound authority, single-use authorization consumption, atomic verify and commit, time-of-check time-of-use prevention, action substitution detection, destination substitution, replay prevention, stale authorization revocation, policy generation epoch, fail-closed unknown state, crash-consistent effect recovery, AI agent action control, agentic re-planning authorization, alternate path closure CPC CLASSES TO CHECK (verify against the current CPC scheme) G06F 21/52 — Monitoring program integrity during execution G06F 21/604 — Tools and structures for managing access control G06F 21/62 — Protecting access to data via a platform H04L 63/10 — Network security, controlling access H04L 63/123 — Verification of received data contents H04L 9/3236 — Authentication using cryptographic hash functions G06N 20/00 — Machine learning
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Recommended Citation
Das, Sangam Kumar, "Method and System for Verifying, at an Effectuation Boundary, That an Operation Proposed by an Autonomous or AI Agent Matches Currently Valid Authority Before Producing an External Effect", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12814