AI in the Build Pipeline: How Software Supply Chain Security Must Evolve

Software supply chain security was already complex, but the integration of AI into the build pipeline has introduced new risks that traditional security programs were not designed to address. For years, the focus was on tracking open-source packages, versions, and transitive dependencies—lessons learned from incidents like SolarWinds, Log4Shell, and XZ Utils. However, the Shai-Hulud malicious package campaign demonstrated that knowing what is in your code is no longer sufficient.

AI tools, models, and infrastructure have become load-bearing components in software development. Code is written by agents, packages are pulled in by autonomous tools, and prompts have become real inputs to the build process—and thus real vectors for compromise. The provenance question now applies to models, agents, and tooling, not just artifacts. An AI coding assistant may suggest a dependency that a developer accepts without human threat modeling, or an autonomous agent may reach for a tool over the Model Context Protocol (MCP) that leads to another tool. Attackers can plant prompts that steer what gets written or pulled in.

To address these challenges, security teams must extend lineage to everything entering the pipeline, including models and agents, and prioritize findings based on real exploitability rather than volume. Gartner formalized this shift in June 2026 with the inaugural Magic Quadrant for Software Supply Chain Security. On July 22, OX researchers will host a webinar titled ‘How AI Is Reshaping Supply Chain Security As We Know It’ to discuss new research, findings from the first systematic look at MCP servers in the wild, and what a supply chain security program looks like when AI is in scope.