Workshops: Secure Agentic Coding & Governance

Software Supply Chain Security for AI-Generated Code: Workshop

Coding agents add dependencies in seconds and often run install commands themselves. Models occasionally suggest packages that do not exist, and attackers register exactly those names. This workshop shows how to secure your software supply chain against these new attack patterns and the familiar ones, without slowing your teams down.

Who is this workshop for?

  • Application security and DevSecOps teams
  • Platform teams running package registries and build pipelines
  • Tech leads and architects responsible for dependencies
  • Open source program offices and licence compliance owners
  • Product security officers at software vendors

What you will take away

  • Explain how hallucinated package names, typosquatting and dependency confusion interact, and spot them in review.
  • Anchor rules in agent configuration so new dependencies are only added after a check.
  • Set up lockfiles, pinning, hash verification and a proxy registry with an allowlist.
  • Generate SBOMs automatically in CI and have agents pre-sort vulnerability findings.
  • Verify signatures and provenance of packages and build artefacts.

Agenda

  1. Module 1

    What AI changes

    • Hallucinated packages and outdated APIs in generated code
    • Agents running install commands on their own
    • Vulnerable patterns reproduced from training data
  2. Module 2

    Attack patterns in the lab

    • Recognising slopsquatting and typosquatting
    • Dependency confusion between internal and public package names
    • Malicious install scripts and compromised maintainer accounts
    • Exercise: systematically verifying an agent's package suggestions
  3. Module 3

    Guardrails for agents

    • Rules files: no new dependency without approval
    • Proxy registry with an allowlist and a cooling-off period for new releases
    • Disabling install scripts or allowing them selectively
  4. Module 4

    Visibility with SBOMs

    • Generating CycloneDX or SPDX SBOMs in the pipeline
    • Software composition analysis and update bots such as Dependabot or Renovate
    • Vulnerability triage with agents, decisions with people
  5. Module 5

    Review and provenance

    • Reviewing lockfile diffs and new dependencies deliberately
    • Signatures and provenance with Sigstore and SLSA
    • Adopting an internal dependency policy

When the agent invents a package

Language models generate plausible package names even when the package does not exist. Research has shown that such names recur, with frequency varying by model and ecosystem. Attackers can register exactly these names on public registries and fill them with malicious code.

If an agent then runs the installation itself, the malicious code may execute before any human has seen the diff. That is why we intervene before installation, not only at review.

A policy for new dependencies

Together we draft a lean policy you can carry over into rules files, pipelines and review checklists.

  • Does the package exist, who maintains it, how active is it?
  • Does the licence fit your open-source policy?
  • Is the version pinned and recorded with a hash in the lockfile?
  • Is there a leaner alternative, or does the standard library suffice?

Ecosystems and regulation

We work with the package managers in your stack, such as npm, pnpm, PyPI, Maven, Gradle, NuGet, Go modules or Cargo, and with coding agents such as Claude Code, OpenAI Codex or GitHub Copilot. Customers and regulations such as the Cyber Resilience Act increasingly expect SBOMs and orderly vulnerability handling; we explain the requirements, while the legal assessment stays with your compliance team.

Frequently asked questions

How often do agents suggest packages that do not exist?

It depends heavily on model, language and task, and published figures vary. What matters is that it happens regularly and can be caught with simple checks before installation.

Should we forbid agents from installing packages altogether?

For many teams a proxy registry with an allowlist plus approval for new dependencies is enough. In sensitive environments a general ban can make sense.

Does an SBOM replace dependency review?

No. An SBOM makes visible what is included. Whether a dependency is trustworthy and necessary remains a deliberate decision in review.

Workshop

Get in touch

Do you know which packages your coding agents have added in recent weeks?

Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.

Prefer to talk first?

15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.

Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)

Request a training / workshop

Workshop: Software Supply Chain Security for AI-Generated Code: Workshop · 1 day

Dates that work for you (optional)

Suggestions about 2 weeks ahead – pick as many as you like

Reply usually within 24 hours · jocapps is ISO 27001 certified · Privacy