Human-Rights Assessment for Agentic Processes: Evidence-Governed Operationalization of Selected Provisions of UN Human Rights Council Resolution 63/15
Year: 2026
Type: preprint
Abstract: An operational record can establish that a human approved an AI-supported decision without settling whether oversight was meaningful or a remedy effective. This conceptual contribution applies a previously proposed evidence-governed Compliance Assessment Framework to selected provisions of UN Human Rights Council Resolution 63/15, adopted in October 2026. It develops an inspectable mapping informed by paragraphs 8, 9, 10 and 13, comprising 22 author-defined assessment elements that retain source addressees, operative force and open qualifiers. It introduces a method-relative evidenceability classification distinguishing directly assessable operational predicates (E1), partially assessable mixed predicates (E2) and judgment-dependent conclusions (E3), with separate atomic subclaims and source-to-target traceability. Five deliberately synthetic recruitment cases illustrate bounded assessment outcomes without asserting real-world reliability, normative fidelity or human-rights compliance. The paper also discusses an optional enterprise governance pathway, including rights-holder-sensitive impact assessment, human oversight by design and effect verification. Grounded in the final English resolution text, the proposed mapping uses a computer-science and information-security perspective to distinguish source conditions from author-proposed operational predicates and evidence requirements while preserving unresolved legal and human-rights qualifiers. It neither establishes compliance, remedy effectiveness, nor a validated control standard, and it is not a checklist of directly enforceable obligations. Its interpretive adequacy and coverage, classification reliability, assessor independence and practical usefulness remain subjects for independent normative and empirical evaluation, as outlined in the Future Work section.
Author: Brömme, Arslan
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7226030632
