AI & DevOps · Supply chain
Supply chain security and regulation, in reading order
The articles on this site about what enters your build and what the law now expects of it, arranged as one path: poisoned packages first, then models and datasets, then what agents commit, then the regulations with dates attached.
By Matthews WongReviewed on
Key takeaway
This path starts with signing, provenance and dependency hygiene, then the npm and PyPI attacks that tested them, treats models and datasets as code that needs signing and scanning, covers the secrets and generated code that AI agents add to a repository, and ends with the Cyber Resilience Act, the EU AI Act, NIST's agent standards work and, for teams in Indonesia, PP 33/2026.
Reading path · 23 articles
Read in this order
Each stage builds on the one before. If you already know a stage, skip ahead; the line under each title says what that article adds.
Stages 1 / 4
Worms and poisoned packages
The foundations of a trusted build, then the npm and PyPI attacks that tested them.
- 01Security10 min readSoftware Supply Chain Security: SBOMs and SigstoreWhy this step: The foundation: generate SBOMs, sign artifacts with Sigstore and cosign, and verify SLSA provenance in CI.
- 02Web Dev9 min readnpm Dependency Management and Security Auditing: What I Actually DoWhy this step: The everyday habits: audit npm dependencies, manage update debt and limit your supply chain exposure.
- 03Security12 min readLiteLLM Supply Chain Attack: PyPI Dependency CooldownsWhy this step: How two malicious releases reached PyPI, and the cooldown, hash and publisher controls that matter.
- 04Security12 min readTensorlake npm Shai-Hulud Compromise: Lock Down Install ScriptsWhy this step: A worm in an npm SDK ran through a preinstall hook; why revoking tokens can backfire and how to block install scripts.
- 05Security11 min readHijacked AI Coding Assistant Session Spread Shai-Hulud WormWhy this step: The worm's spread traced back to a hijacked AI coding assistant session, with the defender steps.
- 06Security12 min readnpm Worm With Valid SLSA Provenance: The keyv LessonWhy this step: A worm shipped with valid provenance: what provenance proves, what it cannot, and the release gates to add.
- 07DevOps11 min readnpm Trusted Publishing: 48-Hour Expiry and issue_comment BanWhy this step: npm now expires unvalidated trusted-publishing setups and rejects issue_comment tokens; a release migration guide.
- 08Security12 min readSlopsquatting in 2026: Package Hallucination Rates, 127 NamesWhy this step: Models invent package names that attackers can register; existence, age and hash checks before you install.
Stages 2 / 4
Models and datasets are artifacts too
Signing, scanning and documenting what you download from a model hub, and why loading it can run code.
- 09Security12 min readOpenSSF Model Signing: Sign and Verify ML Models With OMSWhy this step: Sign and verify model files with OpenSSF Model Signing, and where serving-side enforcement still lags.
- 10Security12 min readPickle Model Scanner Evasion: ShadowPickle and SafetensorsWhy this step: Why a clean pickle scan proves little, and a safetensors-first loading policy.
- 11Security11 min readTransformers trust_remote_code Vulnerability: CVE-2026-4372Why this step: A crafted config.json bypassed trust_remote_code; what the fix changed and a model-loading checklist.
- 12Security12 min readHugging Face Security Incident 2026: Treat Datasets as CodeWhy this step: A dataset loader gave code execution: treat datasets as code, and the rotation order to follow.
- 13Security12 min readAIBOM Hugging Face Study: 97,940 Models, Docs MissingWhy this step: AI bills of materials come out structurally complete but thin on model-card detail; what to add by hand.
Stages 3 / 4
Secrets and code that agents add
The credentials and generated code that end up in a repository, and the checks that catch them in the pipeline.
- 14Security12 min readGitGuardian Secrets Sprawl 2026: AI Agent Commits, MCP ConfigsWhy this step: Secrets leaking through AI agent commits and MCP configs, and how to scan developer endpoints.
- 15DevOps10 min readSecret Scanning and Preventing Credentials Leaks in GitHubWhy this step: Keep credentials out of git history with pre-commit hooks, GitHub secret scanning and gitleaks.
- 16Security11 min readAI Code Security Report 2026: Veracode Stuck at 56% Pass RateWhy this step: How often AI-generated code fails security tests, which weakness classes fail, and how to gate it.
- 17Security11 min readProject Glasswing Patch Gap: 6,157 Disclosed, 516 PatchedWhy this step: Disclosed vulnerabilities far outnumber the patched ones on Anthropic's dashboard; what maintainers and dependents should change.
- 18DevOps10 min readDevSecOps: Integrating Security into CI/CDWhy this step: Wire the checks into the pipeline: SAST, DAST, dependency, image and secrets scanning in CI/CD.
Stages 4 / 4
Regulation and standards with dates attached
What the CRA, the EU AI Act, NIST and Indonesia's data-protection rules ask of a DevOps team, each read in its own article.
- 19Security12 min readCyber Resilience Act Reporting Obligations: A DevOps RunbookWhy this step: The Cyber Resilience Act's reporting clocks as a DevOps runbook, open-source dependencies included.
- 20Security12 min readEU AI Act Digital Omnibus Deadlines: A DevOps Team CalendarWhy this step: Which EU AI Act dates moved and which still apply, laid out as a team calendar.
- 21Security12 min readNIST AI Agent Standards Initiative: What CAISI CoversWhy this step: What NIST's AI agent standards work covers, including agent identity, and what engineers can do now.
- 22Security12 min readPP 33/2026 Pelindungan Data Pribadi: A Developer GuideWhy this step: For teams in Indonesia: PP 33/2026 read against UU 27/2022, including its breach notice and AI rules.
- 23Security13 min readUU PDP Compliance Checklist for AI Features in IndonesiaWhy this step: A UU PDP checklist for AI features: lawful basis, cross-border prompts, provider log retention and deletion.
The AI & DevOps field guide
More from the AI & DevOps guide
The other reading paths in this section.
- MCP & Agent SecurityPrompt injection, the 2026 MCP incidents, access controls, sandboxes and vetting agent skills.
- LLM Inference ServingvLLM and SGLang, disaggregated serving, llm-d and KServe, autoscaling and cost per token.
- Kubernetes for AIThe 1.36 and 1.37 upgrades, DRA and gang scheduling, Gateway API routing and agents in the cluster.
- AI SRE & Agentic OpsIncident-investigation agents, MCP servers for ops tools, agents in CI, and the IaC beneath.