A pre-registered study of 1,920 controlled trials found that coding assistants checked provenance before installing software in just 9 cases, or 0.5%, and never once ran a verification command. The September 2026 research tested whether assistants would use supply-chain signals such as SBOMs, signed releases, build provenance, and official channels when selecting dependencies. The finding points to a shift in how software enters applications: AI agents can now identify dependencies, modify manifests, invoke package managers, and resolve conflicts as part of completing a development task, moving much of the selection process inside an automated workflow. By Cybersecurity Insiders.
Consider a developer asking an agent to connect a Python service to an external API. The agent may add the requests package, which currently has four runtime dependencies — certifi, charset-normalizer, idna, and urllib3. The resulting dependency graph can include five separately maintained packages, each with its own release history and security properties. The organization may prefer a different implementation due to provenance, maintenance history, licensing, or internal security policies, but those considerations are unlikely to be captured in a prompt such as “connect this service to an API.”
Software selection becomes harder to control when an agent handles much of the implementation. A developer may specify what the software needs to do, then rely on the agent and package manager to choose dependencies. The final graph may reflect what the model or ecosystem favors without accounting for every organizational requirement. Good read!
[Read More]