Main article

Isabel Navarro
Department of Information Systems, University of Murcia, Murcia 30100, Spain
Miguel Ángel Rojas
Department of Business Organization and Marketing, University of Vigo, Vigo 36310, Spain
Clara Benítez*
Department of Computer Science, University of Castilla-La Mancha, Albacete 02071, Spain
clara.benitez@uclm.es

DOI: https://doi.org/10.63646/jbda.2023.010405

Abstract

AI-generated security patches are increasingly embedded in software engineering workflows, yet many enterprise release decisions still rely on narrow validation signals such as successful compilation, functional-test passage, static-warning disappearance, or blocking of a single demonstrated exploit. These signals reduce local uncertainty but do not necessarily measure whether a patch removes the vulnerability root cause, resists exploit variants, and preserves business-critical services after release. This article develops a business risk analytics framework for evaluating AI-generated security patches in enterprise software workflows. The framework translates cross-oracle divergence into business risk indicators by linking patch-level validation outcomes to release exposure, remediation delay, operational interruption, compliance risk, and security debt accumulation. A controlled enterprise-style analytical experiment is constructed with 480 AI-generated candidate patches across four vulnerability families: SQL injection, path traversal, cross-site scripting, and missing authorization. Each patch is evaluated across six validation layers: build validity, functional preservation, original proof-of-concept blocking, exploit-variant resistance, root-cause conformance, and regression-safety review. The results show that weak-oracle acceptance substantially overstates business-safe release readiness. In the constructed dataset, 76.7% of patches block the original exploit evidence, whereas only 47.7% satisfy the release-acceptance protocol after variant, root-cause, and regression checks. Oracle divergence is highest for path traversal and missing authorization, where business risk arises from exploit-specific blocking and incomplete trust-boundary repair. The article contributes a managerial analytics model that separates apparent technical repair from release-grade assurance, demonstrates how patch-level validation data can be converted into portfolio risk metrics, and provides governance guidance for enterprises adopting LLM-based vulnerability repair in CI/CD and DevSecOps environments.

Article details

How to Cite

Navarro, I., Rojas, M. Ángel, & Benítez, C. (2023). Business Risk Analytics for AI-Generated Security Patches: Measuring Oracle Divergence in Enterprise Software Workflows. Journal of Business and Data Analytics, 1(4), 94-113. https://doi.org/10.63646/jbda.2023.010405