Skip to content
Halo by OmniTrust

AI governance glossary

AI runtime security

AI runtime security applies controls while AI agents or model requests execute, including identity, policy, network, credential, and action boundaries.

Why it matters

A deployed agent can interact with models and tools while carrying out work. Runtime controls help enforce the boundaries selected for that execution.

How it works

A runtime can isolate workloads and evaluate requests, credentials, and actions as they occur. Its protections depend on the controls implemented and paths in scope.

How Halo approaches it

Halo’s secure agent runtime uses sandboxed, non-root workloads, signed configuration, explicit resource boundaries, and default-deny network egress, with operator quarantine and release controls.

Put a halo around the AI your business runs.

See discovery, runtime policy, approvals, and evidence working together on your own use cases.

Book a walkthrough Start with Halo