Contact Us
Contact Us
Cyber Security

Software Supply Chain Security: SBOMs, Dependency Risks and AI-Suggested Packages

Most of your code is someone else's. Practical controls for SBOMs, dependency scanning, build integrity and packages invented by AI coding tools.

JRC Software Team
6 min read
Software Supply Chain Security: SBOMs, Dependency Risks and AI-Suggested Packages

A critical advisory drops on a Tuesday morning for a logging library you've never heard of. By lunch, a customer emails asking whether you're affected. Your engineers start grepping lock files across forty repositories, and by Thursday they're still not sure about the mobile app. That gap between the question and a confident answer is what software supply chain security is meant to close.

A typical web or mobile project pulls in hundreds of open source packages, most of them transitive, each with its own maintainers and release habits. Supply chain security means knowing what those components are, whether they carry known risk, and whether what you deploy is what your pipeline actually built.

AI coding assistants add a newer problem. They suggest packages, and sometimes the package they suggest doesn't exist.

What changed in 2025 and 2026

First, the SBOM baseline moved. The US Cybersecurity and Infrastructure Security Agency's SBOM page defines an SBOM as "a nested inventory, a list of ingredients that make up software components". In 2026 CISA and international partners, including ASD's ACSC, published 2026 Minimum Elements for a Software Bill of Materials, which ASD published on cyber.gov.au on 8 September 2026. It updates and replaces the 2021 NTIA minimum elements that most tools and buyers had been working from.

Second, AI-generated code introduced a new failure mode. A paper presented at the 34th USENIX Security Symposium generated 576,000 code samples with 16 code-generating models and found the average share of hallucinated packages was at least 5.2% for commercial models and 21.7% for open-source models, with 205,474 unique hallucinated package names. The authors describe package hallucination as persistent and systemic. Attackers can register those names and wait, a technique often called "slopsquatting".

What a 2026-era SBOM should contain

The 2026 guidance splits the minimum elements into data fields and practices and processes. The data fields come in two groups:

GroupElementsWhy you care
SBOM metadataAuthor, author signature, data format name and version, generation context, timestamp, tool name and version, SBOM versionTells a consumer who produced it, with what, at which lifecycle stage, and whether it's been tampered with
Component dataProducer, name, version, identifiers, dependency relationship, hash value and algorithm, licenceLets tools match components to advisories and confirm you're looking at the exact artefact

The new fields compared with 2021 include the author signature, generation context (before build, build or after build), tool name and version, component hash and licence. The practices side covers things like coverage, update frequency, distribution and explicitly marking unknowns rather than leaving gaps.

The practical upshot hasn't changed: generate the SBOM automatically during the build, in a machine-readable format such as CycloneDX or SPDX, for every release. A spreadsheet maintained by hand is wrong the day after someone writes it.

Software supply chain security controls that work

You don't need a large security team for most of the value. Roughly in order of return on effort:

  1. Generate an SBOM on every build. Tools like Syft or the CycloneDX plugins for npm, Maven and Gradle do this in one CI step. Store it with the release so "are we affected?" becomes a search.
  2. Scan dependencies on every pull request and nightly. Software composition analysis tools such as OWASP Dependency-Check identify dependencies and check for known, publicly disclosed vulnerabilities. Nightly runs catch advisories published against code you haven't touched.
  3. Commit lock files and install from them. Use npm ci rather than npm install in CI, so a surprise minor release can't slip into a build.
  4. Automate updates, then review them. Dependabot or Renovate keeps you current. Agree how fast security updates get merged and make sure tests run on every update pull request.
  5. Pull through a registry you control. A proxy such as Artifactory, Nexus or a cloud artifact registry lets you block known-bad packages and keeps a cache if an upstream package vanishes.
  6. Check the health of key dependencies. OpenSSF Scorecard runs automated checks on open source projects, including branch protection, code review, pinned dependencies, token permissions and signed releases, and scores each out of 10.
  7. Protect the build itself. The SLSA build levels run from no guarantees, through provenance that records how a package was built, to builds on a hardened platform that resists tampering. SLSA's site marks v1.0 as retired and v1.2 as current, so read the current spec before you set a target level.
  8. Sign artefacts and verify at deploy. Only deploy images and packages your pipeline built and signed.

When the assistant picks the package

With a hallucinated dependency, nobody chose it in any meaningful sense. The assistant did, and the developer accepted a plausible-looking import. So the control has to sit on the dependency, not on the person.

What works for teams using AI coding tools:

  • Make new dependencies visible. Add a CI check that lists new entries in the lock file and posts them on the pull request, so "adds a package" can't hide inside a 600-line diff.
  • Verify before approving. Look at the registry page, the source repo, maintainer history and download trend. A brand-new package with a plausible name, no repo and a few dozen downloads should get a hard look.
  • Block installs from outside your registry. If everything comes through your proxy, a hallucinated name that someone has since registered can be quarantined centrally.
  • Prefer what you already use. Put "use existing dependencies where possible" in your team's AI coding guidelines and in the assistant's project instructions.
  • Scan AI-heavy codebases more often. Apps built fast with AI tools pile up dependencies fast. Our guide to testing AI-generated code covers the wider quality checks.

Where this fits in your wider security programme

Supply chain controls connect to the baselines you probably already report on. Patching applications is one of the Essential Eight strategies, and in a modern codebase that mostly means dependencies, as we explain in Essential Eight for software teams. Design matters too: a threat modelling workshop is a good place to decide which third-party services and packages get trusted with sensitive data.

Mobile apps need the same attention, since third-party SDKs often collect data or request permissions you didn't plan for. Our mobile app security checklist covers those release checks.

How JRC Software helps

Our secure development and DevSecOps team starts with a pipeline security scan. In the first week we generate SBOMs for your main applications, run dependency and container scans, trace how packages enter your builds (straight from public registries, or through a proxy) and list the permissions your pipeline tokens hold.

You get a short findings report ranked by risk, not a 3,000-row tool export. We'll also tell you what to skip for now: chasing SLSA level 3 makes little sense while you're still running npm install against the public registry. If you want help with the fixes, we add SBOM generation, sensible scan thresholds, a new-dependency check for AI-assisted work and artefact signing alongside your engineers.

If you want to know what's really inside your software, request a pipeline security scan.

Frequently asked questions

What is software supply chain security?

It's managing the risk from everything your software depends on or passes through before reaching users: open source packages, third-party SDKs, build tools, CI/CD pipelines and container images. The aim is to know what you ship, keep it free of known vulnerabilities, and prove the artefact you deploy is the one your trusted pipeline built.

Do small teams really need an SBOM?

Yes, and it's cheap. Most build ecosystems have tools that generate one in a single CI step, so the cost is mostly setup. The payoff is speed: when a serious advisory lands, you search your SBOMs instead of inspecting every repository. Enterprise buyers also increasingly ask for SBOMs during procurement and security reviews.

What is slopsquatting?

Slopsquatting is registering package names that AI coding tools tend to invent, so developers who accept the suggestion install the attacker's code. Research presented at USENIX Security 2025 found hallucinated package names are common across code-generating models and describes the problem as persistent and systemic. Reviewing every new dependency and installing only through a controlled registry are the best defences.

How is dependency scanning different from an SBOM?

An SBOM is an inventory of the components and versions in a release. Dependency scanning compares that inventory, or your manifests and lock files, against vulnerability databases and flags known issues. You need both: the SBOM tells you what you have, and scanning tells you which of those components carry known risk today.

Cyber Security

Request a pipeline security scan

Tell us where you're stuck and we'll come back with practical next steps.

Get in touch

Sources