Safeguard
Supply Chain•

How do I vet a new AI library before adding it to my stack?

A 15 minute vetting checklist for new AI and agent libraries: maintainers, release habits, advisories, install behavior, license, and how deep it reaches.

Safeguard Security Team
3 min read

Vet a new AI library by checking who maintains it, how it releases, what advisories it has, what it does at install time, which license it carries, and how many transitive packages it adds. Fifteen minutes on these six checks catches most avoidable problems before they enter your lockfile.

Why AI libraries deserve a closer look

AI tooling moves fast. New agent frameworks, connectors, and wrappers appear weekly, often from small teams, and they tend to request broad access: your model keys, your files, your tools. OWASP's supply chain entry for LLM applications lists vulnerable packages, unclear licenses, and weak provenance among its concerns.

The six checks

CheckQuestionWhere to look
MaintainersMore than one active maintainer? Is the account protected?Repository and registry page
Release habitsRegular tagged releases, changelog, signed or attested builds?Release page
AdvisoriesPast advisories and how fast they were fixedAdvisory databases
Install behaviorInstall scripts, native build steps, network calls at installPackage contents
LicenseCompatible with how you ship it?License file and metadata
DepthHow many transitive packages does it add?Dependency tree

Use a scorecard as a starting point

OpenSSF Scorecard runs 18 automated checks on open source projects, including maintenance status, security policy, dependency update tooling, pinned dependencies, code review, branch protection, and signed releases. It can be run from the command line on any project. A low score is a prompt to look closer, and a high score is not a guarantee, but it saves you from collecting the basics by hand.

Check the advisory history, not just today's list

Search the GitHub Advisory Database and OSV for the package name. Note how long fixes took after reports. Zero advisories for a young project can mean no scrutiny, not good health.

Decide how much access it gets

Run unfamiliar agent or connector code with minimal privileges: no production keys, a scoped network, a throwaway workspace. If it needs broad permissions to work, treat that as part of the decision.

Record the decision

Write down why you accepted it and who owns upgrades. Pin the version, and review each update before taking it.

A short note on Safeguard

Safeguard offers SBOM and dependency scanning, plus asset discovery, so you can see what a new library added to your inventory. Try a free scan at gold.safeguard.sh/scan.

Quick check

Is a high star count a sign of safety?

No. Popularity helps you find problems faster but does not measure maintenance or security practice.

Should every dependency get this review?

Do the full review for libraries that touch keys, files, or tools, and a quick scan for the rest.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.