Shift from auditing everything to knowing everything
You cannot realistically review every line of every transitive dependency. You can maintain an accurate, automatically generated software bill of materials, so that when a vulnerability is disclosed somewhere in the ecosystem, you can answer 'are we exposed' in minutes rather than days of manual investigation.
This single capability — knowing what you actually run, accurately and automatically — is the highest-leverage supply-chain investment most organisations can make, and it's achievable without reading a single line of the dependencies themselves.
Pin, verify, and control the update path
Pin dependency versions deliberately rather than always-latest, verify package signatures where the ecosystem supports it, and route updates through the same review process as any other code change rather than letting automation merge them unreviewed.
This doesn't mean rejecting automated dependency updates — it means routing them through a gate that at minimum runs your existing test and security scanning suite before merge, the same as a human-authored change.
Extending the same discipline to AI supply chains
The newest version of this problem is model weights, fine-tuning datasets and MCP tool servers sourced from third parties — each one is a supply-chain dependency with provenance and integrity questions, even though it doesn't look like traditional code.
Apply the same bill-of-materials discipline here: know exactly which model versions, data sources and tool servers are in production, and treat a change to any of them with the same review rigor as a dependency update.
You cannot audit a dependency tree you cannot even enumerate. Start by knowing exactly what you run.MNB Security Practice
Key actions
- Generate an accurate, automatic software bill of materials
- Pin dependency versions and verify signatures where supported
- Route all dependency updates through existing review and test gates
- Apply the same provenance discipline to model weights and MCP tool servers