Glossary
Glossary
This glossary defines the deployment and governance terms used to describe Udal and AI agents in regulated organisations.
Sovereign AI
Sovereign AI is an approach to AI in which an organisation controls where processing happens and which rules apply to its data and models. It depends on deployment, access and operational controls, not simply the location of a supplier.
AI agent control plane
An AI agent control plane is software that governs what agents can access, which models they use and what they spend. It applies policy, approval and audit controls across agent activity.
Model routing
Model routing selects a model for a request according to rules such as sensitivity, capability and cost. It can keep sensitive work on internal models and permit external providers only where policy allows.
Open-weight model
An open-weight model makes its trained parameters available for others to run. Availability of weights does not necessarily mean the training data, source code or licence is unrestricted.
Air-gapped deployment
An air-gapped deployment runs in an environment isolated from external networks. Software updates, model imports and evidence exports require a controlled transfer process.
Tamper-evident audit log
A tamper-evident audit log records events so that later changes can be detected. It supports investigation and assurance but does not mean records are impossible to alter.
Human-in-the-loop approval
Human-in-the-loop approval requires a person to review an action before it proceeds. Policies can apply approval gates to sensitive data access, spending or changes to business systems.
Software bill of materials
A software bill of materials (SBOM) is an inventory of software components included in a release. It helps teams review dependencies, licences and known vulnerabilities.