Newest papers

Reusing Assurance Evidence in AI Pipelines: Validity Conditions, a Fail-Closed Release Gate, and a Replay of 714 Public Data Version Control Pipelines

Alka Mukhopadhyay, Shravya Reddy Nalla · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.27672 · cited by 0 works (OpenAlex) · open-access PDF

French

Le renouvellement de chaque contrôle d'exposition après chaque changement dans un pipeline d'intelligence artificielle (IA) est coûteux, de sorte que les équipes réutilisent les résultats antérieurs. Le renouvellement n'est sûr que si les dépendances déclarées de chaque contrôle couvrent tout ce qui peut changer son résultat, et seulement si les vérifications de la porte de libération qui réutilisent les preuves décrivent le candidat qui est libéré. Cet article fait les deux conditions explicites et mesure combien de fois le premier échoue en pratique. Il définit quand un enregistrement de preuve signé est valide pour un candidat et donne des conditions suffisantes dans lesquelles la réexécution de changement local atteint la même décision de libération que le renouvellement de chaque contrôle. Une porte de référ

Machine translation (Meta M2M100); the English original follows.

Abstract

Rerunning every governance control after every change to an artificial intelligence (AI) pipeline is expensive, so teams reuse earlier results. Reuse is safe only if each control's declared dependencies cover everything that can change its outcome, and only if the release gate checks that reused evidence describes the candidate being released. This article makes both conditions explicit and measures how often the first fails in practice. It defines when a signed evidence record is valid for a candidate and gives sufficient conditions under which change-local re-execution reaches the same release decision as rerunning every control. A reference gate implementing these checks with Ed25519 digital signatures blocks eight of nine manipulation scenarios, against one for a gate that trusts the planner and three for a gate that only verifies signatures; a dependency the policy never declared passes every gate. A replay of 38,083 commits from 714 public Data Version Control (DVC) pipelines measures that failure mode. Reusing stage results according to declared dependencies avoids 83.4% of stage executions, but in 3.3% of commits it reuses a stage whose imported code or referenced configuration changed, and 31% of pipelines are affected. A manual audit confirms the undeclared dependency in 36 of 40 sampled events. Evaluation stages, whose outputs are the evidence, are the most exposed. Adding each stage's static import closure lowers stale reuse to 0.4% while still avoiding 81.6% of executions; sampled shadow re-execution removes far less, and rerunning everything after unexplained changes avoids only 45.3%.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Reuse, Dependency (UML), Pipeline transport, Pipeline (software), Computer science, Control (management)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange