INDEPENDENT DESIGN PROPOSAL · VERSION 1.0.0

PICEE-PA

Governed AI Execution for Public Administration

Policy → Identity → Capability → Execution → Evidence

AI-generated intention is not authority to act. Significant actions should be explicitly authorized, bounded, enforceable and reconstructible.

Contribution

PICEE-PA does not claim to invent policy engines, agent identity, sandboxing or audit. It proposes a vendor-neutral public-administration application profile that connects administrative authorization — purpose, process, delegation and accountability — to technical authorization — identity, capability, policy enforcement and runtime controls — for each AI-initiated action.

Architecture

PICEE-PA governed AI execution architecture showing Policy, Identity, Capability, Execution and Evidence

Five layers

Policy

Purpose, process, applicable rules, constraints and risk.

Identity

Requester, agent or workload, sponsor and delegation chain.

Capability

Bounded action, resource, data scope and budgets.

Execution

Fail-closed enforcement, approvals, isolation and recovery.

Evidence

Attributable policy decision, actual action, outcome and provenance.

AEE and AEB

AEE — Administrative Execution Envelope is the proposed structured pre-execution envelope describing the administrative and technical authority under which an AI-initiated action may be considered.

AEB — Administrative Evidence Bundle is the proposed post-execution evidence package supporting durable reconstruction of authorization, execution, approvals, outcomes and provenance.

Downloads and citation

DOI 10.5281/zenodo.22229330
Permanent citation

Lombardi, V. (2026). PICEE-PA: Governed AI Execution for Public Administration . Version 1.0.0. Zenodo.
DOI: 10.5281/zenodo.22229330

Author

Valerio Lombardi · Computer Engineer · Independent Researcher

ORCID: 0009-0006-3949-6228

Independent personal research. The views expressed are solely those of the author and do not represent any organisation, institution or employer.