Delegation Under Evidentiary Drift

Automated systems increasingly exercise delegated authority under predefined rules and authorization controls. Those rules may remain unchanged while the quality, meaning, provenance, completeness, or applicability of the evidence used to satisfy them changes. This preregistered study examines whether material degradation in evidentiary warrant can create a mismatch between apparently valid delegated authority and the evidence supporting its continued exercise. The study distinguishes two failure modes: detection failure, in which material evidentiary degradation remains undetected while delegated execution authority remains available, and constraint failure, in which degradation is detected but the prespecified reevaluation or authorization-state response does not occur within the registered observation window. Detection and constraint are evaluated separately rather than treating awareness of degraded evidence as equivalent to effective governance. The experimental design evaluates preregistered classes of evidentiary degradation against a prespecified baseline of representative authorization and change-control functions. Evidentiary degradation is introduced under controlled conditions, providing experimental ground truth independent of the controls being evaluated. Before experimental execution, the completed restricted specification fixes the materiality criteria, drift classes, control conditions, observation windows, repeated-run decision rules, exclusion criteria, and outcome-classification criteria that govern the experiment. Mixed outcomes are retained and reported rather than retrospectively collapsed into a favorable binary interpretation. Cases involving prior-authorization adjudication, automated or AI-assisted patching, governed change, and human exception handling motivate and discriminate among related propositions but do not independently establish the primary hypothesis. A negative comparison condition is included to demonstrate detection under a prespecified condition. The study does not assume that AI uniquely creates the problem, that automation is inferior to human judgment, or that evidentiary drift necessarily produces unsafe action. Its narrower objective is to determine whether, and under which preregistered conditions, evidentiary warrant can deteriorate while apparently valid delegated authority persists. This public preregistration records the research question, hypothesis structure, outcome taxonomy, case propositions, mechanism taxonomy, interpretation rules, and claims boundary before experimentation. A separately versioned restricted record fixes the concrete experimental design before the first experimental run while maintaining the study's disclosure boundary.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24374312
Primary Topic
Access Control and Trust
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Delegation Under Evidentiary Drift

Narnaiezzsshaa Truong
Zenodo (CERN European Organization for Nuclear Research)
Access Control and Trust
article

Delegation Under Evidentiary Drift

Narnaiezzsshaa Truong
article en

Abstract

Automated systems increasingly exercise delegated authority under predefined rules and authorization controls. Those rules may remain unchanged while the quality, meaning, provenance, completeness, or applicability of the evidence used to satisfy them changes. This preregistered study examines whether material degradation in evidentiary warrant can create a mismatch between apparently valid delegated authority and the evidence supporting its continued exercise. The study distinguishes two failure modes: detection failure, in which material evidentiary degradation remains undetected while delegated execution authority remains available, and constraint failure, in which degradation is detected but the prespecified reevaluation or authorization-state response does not occur within the registered observation window. Detection and constraint are evaluated separately rather than treating awareness of degraded evidence as equivalent to effective governance. The experimental design evaluates preregistered classes of evidentiary degradation against a prespecified baseline of representative authorization and change-control functions. Evidentiary degradation is introduced under controlled conditions, providing experimental ground truth independent of the controls being evaluated. Before experimental execution, the completed restricted specification fixes the materiality criteria, drift classes, control conditions, observation windows, repeated-run decision rules, exclusion criteria, and outcome-classification criteria that govern the experiment. Mixed outcomes are retained and reported rather than retrospectively collapsed into a favorable binary interpretation. Cases involving prior-authorization adjudication, automated or AI-assisted patching, governed change, and human exception handling motivate and discriminate among related propositions but do not independently establish the primary hypothesis. A negative comparison condition is included to demonstrate detection under a prespecified condition. The study does not assume that AI uniquely creates the problem, that automation is inferior to human judgment, or that evidentiary drift necessarily produces unsafe action. Its narrower objective is to determine whether, and under which preregistered conditions, evidentiary warrant can deteriorate while apparently valid delegated authority persists. This public preregistration records the research question, hypothesis structure, outcome taxonomy, case propositions, mechanism taxonomy, interpretation rules, and claims boundary before experimentation. A separately versioned restricted record fixes the concrete experimental design before the first experimental run while maintaining the study's disclosure boundary.

Zenodo (CERN European Organization for Nuclear Research)
Access Control and Trust
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon