Authority Objects for AI-Triggered Execution: A Formal Specification for Machine-Verifiable Legitimacy in the LPP Admission Kernel

This preprint introduces Authority Objects as machine-verifiable legitimacy carriers for AI-triggered execution under the Lingua Pactum Protocol (LPP). As AI systems transition from language generation to tool use, API invocation, workflow automation, financial execution, cyber operations, enterprise infrastructure, robotic systems, and other external action environments, governance cannot rely solely on model confidence, runtime access control, human-facing explanations, or post-hoc audit. A system identity may have permission to call an API, and a model may provide a plausible explanation, while the specific AI-triggered action may still lack legitimate authority. This paper formalizes Authority Objects as the upstream legitimacy substrate of the LPP Admission Kernel. An Authority Object binds subject identity, issuer identity, action scope, jurisdictional scope, irreversibility or risk class, signature requirements, signature state, validity window, temporal legitimacy state, revocation state, policy binding, cross-system recognition, consensus evidence, and lifecycle status into a structured object evaluated before execution. The paper defines Authority Object structure, lifecycle states, minting semantics, revocation behavior, bounded temporal degradation, cross-system transfer constraints, relationship to Execution Permits, relationship to Admission Artifacts, failure modes, governance invariants, and comparisons with IAM, access tokens, capability tokens, Macaroons, policy engines, consent records, audit logs, and human approval workflows. This work is part of the public LPP research chain. It follows the LPP constitutional admissibility paper, the LPP Admission Kernel operational semantics paper, and the LPP-FIN domain application paper. Its central claim is that AI-triggered execution requires a legitimacy object that is external to the model, scoped to the action, lifecycle-bound, revocable, and machine-verifiable before execution. This work is released under CC BY-NC 4.0 for academic citation, sharing, and non-commercial discussion.