AI and LLM Security
AI Supply Chain Security
We assess the origin of the models, datasets, plugins and libraries you use, and whether they have been altered. We show what an external component could introduce into your system.
What it is
AI systems rely on models trained and datasets prepared by others. Some model file formats can execute code when loaded. Tampering with training data or a pre-trained model can change the model's behavior without detection. For high-risk systems, the EU AI Act requires measures against data and model poisoning where appropriate.
Scope
- Model source, publisher and integrity verification
- Detection of model file formats that can execute code when loaded
- Origin and tampering risk of training and fine-tuning data
- Dependency analysis of MCP servers, plugins and AI libraries
- Data retention and access terms of model provider APIs
- AI component inventory (AI-BOM)
The need
When organizations need it
- When teams start adding models downloaded from open model repositories to the product.
- When a model will be fine-tuned on a dataset obtained from outside the organization.
- When a corporate customer requests an inventory of the AI components in your product.
Method
How we work
-
Inventory
We list models, datasets, plugins and libraries with their sources.
-
Origin and integrity
We verify each component's publisher, version and hash value.
-
File and permission review
We review model file formats and the permissions requested by plugins.
-
Intake rules
We recommend rules defining which sources new components may come from and which checks they must pass.
Outcome
What you have at the end
You know the origin and verification status of every AI component in your system. The required checks for new components are documented.
Frequently asked questions
Is using open-source models risky?
It depends on the source and format. A model from a verified publisher, stored in a format that does not execute code, carries less risk. A model of unknown origin going directly into the product is a serious risk.
We use a commercial model API. Is there still a supply chain risk?
Yes, but in a different way. You do not load the model, but your prompts and data go to the provider. Contract terms, data retention and model version changes are assessed.
AI and LLM Security
Other services in this area
Let's talk about this service.
Tell us briefly what you need. We will agree on the scope and timing together in the first meeting.
Schedule a Meeting