Main article

Nur Aisyah Rahman
Department of Software Engineering, Universiti Teknikal Malaysia Melaka, Melaka, Malaysia, 76100
Hafizuddin Yusof
Faculty of Computing, Universiti Malaysia Pahang Al-Sultan Abdullah, Pekan, Malaysia, 26600
Mei Wen Tan*
School of Computer and Communication Engineering, Universiti Malaysia Perlis, Arau, Malaysia, 02600
meiwen.tan@unimap.edu.my

DOI: https://doi.org/10.63646/cft.2024.020303

Abstract

Autonomous DevSecOps agents are beginning to change how software organizations discover, prioritize, repair, and release security patches. Yet an agent that produces a plausible patch is not necessarily producing a reliable security fix. A patch may compile, pass unit tests, silence a scanner warning, or block the original proof-of-concept exploit while still failing semantically equivalent exploit variants, missing the vulnerability root cause, or introducing release-level regressions. This article develops a forward-looking analytical framework for cross-oracle validation in AI-driven software supply chains. The proposed framework treats patch validation as a business risk analytics problem rather than a narrow code-correctness problem. It integrates agentic patch proposal, evidence collection, oracle-strength sequencing, exploit-variant testing, root-cause conformance review, regression safety, and release governance into a single decision pipeline. A diagnostic data analysis is conducted on a constructed enterprise workflow benchmark of 72 vulnerability repair tickets and 360 AI-generated patch candidates across six workflow domains. Results show that original exploit blocking accepts 82% of generated patches, whereas full cross-oracle release approval accepts 51%, producing a 31-percentage-point validation gap. Cloud infrastructure-as-code and identity-access workflows show the highest oracle divergence, while dependency upgrade workflows show the strongest regression-risk profile. The analysis further indicates that a cross-oracle agent reduces residual release risk by 37% relative to a single-oracle agent, although it increases median validation delay by 2.8 hours. The article contributes a business-oriented evaluation architecture for future DevSecOps agents and offers governance recommendations for integrating automated patch repair into software supply chain risk management.

Article details

How to Cite

Rahman, N. A., Yusof, H., & Tan, M. W. (2024). Toward Autonomous DevSecOps Agents: Cross-Oracle Validation for Future AI-Driven Software Supply Chains. Crossroads of Future Technologies, 2(3), 38-57. https://doi.org/10.63646/cft.2024.020303