Bottom line: AI coding assistants are pulling open-source packages into development processes at a pace that traditional manual security reviews can no longer cover.
AI-powered coding tools suggest open-source dependencies faster than traditional security checks can vet them. ActiveState points out that many of these packages are unvalidated or simply hallucinated, and enter development pipelines without review.
In its analysis, ActiveState describes a structural problem arising from the widespread use of AI coding tools: these tools generate code containing dependencies on external open-source packages without those packages having first gone through a security or license review process. According to ActiveState, some of these suggested packages don’t actually exist at all — they are a hallucination produced by the respective model, a circumstance that creates additional risk, for instance through so-called slopsquatting, where attackers deliberately register such hallucinated package names in order to inject malicious code.
For CISOs, this shifts the control question: classic Software Composition Analysis (SCA) processes typically only kick in once code is already in the repository or a pull request review is pending. With AI-generated code, however, the dependency problem already arises at the moment of code creation by the developer or the tool itself — well before any established review instance. The volume of newly suggested packages is growing faster than security teams can manually assess them with existing staff resources.
ActiveState therefore recommends shifting package governance forward to the point of selection, rather than only controlling it retroactively in the pipeline. Concretely, this means: curated, already-vetted package catalogs or registries from which AI tools and developers may exclusively source packages, as well as automated control points that block unknown or unverified package names before they ever enter the build process.
For security leaders at organizations actively using AI coding assistants — such as GitHub Copilot, Cursor, or comparable tools — this creates a need to review existing software supply chain policies. Pure after-the-fact checks in the CI/CD process are no longer sufficient once the volume of AI-generated dependency suggestions exceeds the capacity of existing reviewer teams.