Models
Foundation models, open-source weights pulled from public hubs, and fine-tuned variants built in-house.
Every model, dataset, library and vendor behind your AI is part of a supply chain. Here is what it is, where it breaks and how to manage it.
AI supply chain risk management is the practice of identifying, assessing and controlling the risks that enter an organization through the components its AI depends on: third-party and open-source models, training and fine-tuning data, AI libraries and frameworks, infrastructure, and the vendors that embed AI in their products.
It extends software supply chain security to the parts of AI that traditional tools cannot see.
An AI system is assembled, not written. Each link below is a dependency you inherit, and each one can carry risk into production.
Foundation models, open-source weights pulled from public hubs, and fine-tuned variants built in-house.
The datasets used to train, fine-tune, evaluate and ground a model, along with where they came from.
AI libraries, SDKs, agent frameworks and the packages they pull in transitively.
The cloud services, endpoints and pipelines that host, serve and connect AI systems.
SaaS tools and suppliers that embed AI in what they sell you, often without telling you.
Third-party AI risk managementAutonomous agents and the tools, plugins and APIs they are allowed to call on your behalf.
Supply chain risk is inherited risk: you did not create it, but you are accountable for it once it runs in your environment.
| Risk | What it looks like |
|---|---|
| Shadow AI | Models, agents and AI features in use that no one has inventoried or approved. |
| Tampered or unsafe models | Third-party weights or files that behave differently from what was published. |
| Data provenance gaps | No record of what data trained a model, or whether it was licensed and permitted. |
| Vulnerable dependencies | AI libraries and packages with known weaknesses pulled into production. |
| Vendor opacity | Suppliers using AI on your data with no evidence of how it is secured or governed. |
| Compliance exposure | No proof for the EU AI Act, NIST AI RMF or ISO 42001 when auditors ask. |
Cranium AI organizes AI supply chain risk management around five moves, run continuously rather than once a year.
Inventory every model, agent, dataset and vendor, including shadow AI, and record it in an AI bill of materials (AI‑BOM).
Watch how those components behave in production so drift, anomalies and unexpected use surface early.
Turn policy into enforced controls and map each component to the frameworks that apply to it.
Red team models and agents against real adversarial attacks before attackers do.
Share current, framework-mapped evidence with auditors, customers and regulators on demand.
Detect AI, CodeSensor, CloudSensor and AgentSensor find AI across code, cloud and agents and generate AI‑BOMs automatically.
See DiscoverBring vendor AI into the same inventory as the AI you build, so one policy covers both.
See third-party AIArena runs human and AI-driven red teaming against models and agents.
See SecureCranium AI Cards map each system to the EU AI Act, NIST AI RMF and ISO 42001 and can be shared through Trust Hubs.
See ProveStraight answers, kept current. Last updated .
Ask us anything →It is the practice of identifying, assessing and controlling risks that enter through the components AI depends on: models, data, code, infrastructure and third-party vendors.
Software supply chain programs track code packages. AI adds models, training data and vendor AI, which software bills of materials and SBOM tools do not detect. An AI‑BOM closes that gap.
Discovery. You cannot assess what you have not found, so start with an inventory of every model, agent, dataset and vendor, including shadow AI, recorded in an AI‑BOM.
Yes. Vendors that embed AI in their products are part of your AI supply chain, and their risk becomes yours once their AI touches your data or decisions.
The EU AI Act, the NIST AI Risk Management Framework and ISO/IEC 42001 all expect organizations to understand and manage the AI components they rely on.
Cranium ONE discovers AI across code, cloud, agents and vendors, generates AI‑BOMs, red teams models with Arena and proves compliance with AI Cards, following the AI Trust Loop: Discover, Observe, Govern, Secure and Prove.
See how Cranium AI discovers, observes, governs, secures and proves every model, agent and vendor in your enterprise.
We use essential cookies to run this site, and — only with your permission — analytics and marketing cookies to improve it. You're the hero here; you stay in control. Privacy Policy.