AI-Assisted Secure Coding (Appendix C) · Adversarial AI Detection in Inbound Contributions
Detect suspicious dependencies in PRs
Verify that automated PR pipeline gates detect known indicators of LLM-generated or LLM-assisted malicious payload patterns: registry-confusable or typosquatted dependency names, package references that do not resolve to any published version, and dependencies whose creation, first-publication, or maintainer-change timestamps look anomalous relative to the PR.
OWASP AISVS v1.01, requirement AC.13.3
In plain English
Check pull requests automatically for the dependency tricks that AI-assisted attackers use: look-alike package names, references to packages that do not exist, and packages that were created, first published or handed to a new maintainer suspiciously recently.
What good looks like
- Every dependency change in a PR is checked automatically.
- Typosquatted and registry-confusable names are flagged.
- References that resolve to no published version are flagged.
- Unusual creation, first-publication or maintainer-change dates relative to the PR are flagged, and flags block the merge until reviewed.
How teams usually do it
- Dependency review in CI together with a package-reputation service or tool (for example OpenSSF Scorecard data, deps.dev, or a commercial supply-chain scanner).
- A minimum package age rule, such as refusing versions published in the last few days.
- An internal registry proxy that serves approved packages only, and scoped names to prevent dependency confusion.
Suggested evidence
- CI configuration.
- Flagged examples.
- The package age and allow-list rules.
Common mistakes
- Checking known vulnerabilities only, which misses brand-new malicious packages.
- Checking direct dependencies and ignoring lockfile changes.
- Trusting a package name because an AI assistant suggested it.
Not sure which requirements apply to your system? Take the assessment, or browse every guide in this chapter.
Requirement text from the OWASP AI Security Verification Standard, licensed under CC BY-SA 4.0. This guidance is shared under the same licence.