Hallucinated dependencies are fake packages, libraries, or registries that an AI model invents while trying to solve a coding task. They create real security risk because attackers can register those names on public package managers and place malicious code there, turning a model error into a software supply chain compromise.
What Hallucinated Dependencies Are
Hallucinated dependencies are not just a model mistake, they are a supply chain hazard. When an AI invents a package or registry name, it creates a false but plausible target that an attacker can register later, so the error can become a real delivery path for malicious code.
This is distinct from ordinary coding error because the dependency may not exist at all until someone weaponises the name. The security problem is therefore not only incorrect output, but the creation of a believable artefact that can be exploited in package ecosystems.
Why They Are Dangerous in Software Development
The core risk is trust. Developers often treat model suggestions as if they came from a search result or a dependency resolver, but hallucinated names can bypass that assumption and lead to typosquatting-like compromise, poisoned installs, or accidental use of an unreviewed package.
That danger grows when the suggested dependency sits in build scripts, automation, or agent-generated pull requests, because the mistaken reference can move quickly from a suggestion to a production dependency without enough human verification.
In practice, the issue belongs to software supply chain security as much as to AI safety. A hallucinated dependency can become a staging point for malicious code, dependency confusion, or upstream impersonation if organisations do not verify the package source before adoption.
How Organisations Check Exposure
Exposure checks should start with dependency inventory, not model intent. Teams need to know which packages are actually approved, which registries are trusted, and whether the organisation allows direct installation from public ecosystems or only from curated mirrors and internal registries.
That review should be paired with a check of internet-facing build pipelines and automation paths, because an environment that can fetch arbitrary packages on demand is easier to abuse than one that pins sources and validates artefacts. Strong software supply chain controls help reduce the chance that a false package name can reach a live build.
Where code assistants are involved, organisations should also review how generated dependency suggestions are accepted, reviewed, and promoted. A secure workflow should make it obvious when a dependency was model-suggested, because that context changes the level of verification required.
Remediation, Detection, and Post-Compromise Review
The safest response is to reject unverified dependencies by default and resolve them against trusted sources before they are merged or deployed. If a patching or migration window is involved, compensating controls such as allowlisted registries, package provenance checks, and locked manifests reduce the chance of accidental ingestion.
Detection should focus on unusual package introductions, unexpected install sources, and new dependencies that do not match internal software standards. If a hallucinated package was ever installed, the review should extend beyond the code change itself to confirm whether build artefacts, CI runners, or developer environments pulled a malicious payload.
After any suspected exploitation, teams should treat the event like a supply chain compromise: review build logs, dependency locks, package caches, signing material, and any tokens used by automation that could have been exposed during retrieval or installation.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
SLSA, OWASP ASVS, CIS Controls v8 and NIST SP 800-53 Rev 5 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| SLSA | Supply-chain Levels for Software Artifacts | Hallucinated dependencies can introduce untrusted software into builds and releases. |
| Recommendation — Require provenance checks before accepting any model-suggested dependency into the build pipeline. | ||
| OWASP ASVS | V13 — Configuration | Dependency source selection and installation paths are configuration decisions affecting trust. |
| Recommendation — Constrain dependency sources and validate configuration so unapproved packages cannot be installed. | ||
| CIS Controls v8 | CIS-16 — Application Software Security | Controls for secure software development and validation apply to dependency introduction. |
| Recommendation — Verify third-party components before use and block unreviewed dependencies from production builds. | ||
| NIST SP 800-53 Rev 5 | SA-12 — Supply Chain Protection | The term describes a supply-chain integrity failure that this control family directly addresses. |
| Recommendation — Apply supply-chain protection requirements to vet, approve, and monitor external software components. | ||
Practitioner Guidance
Why practitioners should care: Hallucinated dependencies are a governance problem as much as a coding problem, because they can convert a harmless model error into an unsafe procurement decision. Treat every model-suggested package name as untrusted until it is verified against an approved source of truth.
What to watch for: The highest-risk cases are build automation, agentic coding flows, and fast-moving pull requests where nobody independently checks whether the suggested package actually exists. If the dependency is not in your approved catalogue, assume it needs manual review before it enters the software lifecycle.
Practitioner takeaway: The practical control is disciplined dependency validation, not better prompt wording. If the package is real, confirm provenance; if it is not, do not let the model create one for you.
Related resources from NHI Mgmt Group
- Why do insecure training data and hallucinated dependencies increase application risk?
- How should security teams handle hidden AI framework dependencies in enterprise environments?
- What breaks when teams rotate secrets without mapping dependencies first?
- Why do third-party dependencies make auth outages worse?
Deepen Your Knowledge
Free weekly newsletter
Subscribe to the NHI & AI Identity Journal
The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.
Bonus 33% off our NHI Course when you subscribe.
Reviewed and updated by the NHIMG editorial team on October 31, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org