Research Notes
AIDR Runtime Security Notes
Working principles for building an endpoint-first detection, investigation, and response platform for AI agents.
Updated August 16, 2026
Why the endpoint is the first control plane
Agents inherit user privileges and act through local tools, files, credentials, and network access. Endpoint runtime evidence is therefore the earliest practical place to observe behavior and apply validated controls.
Drift is a starting point, not a verdict
A new model, tool, endpoint, or call pattern is a signal for investigation. The decision must return to session evidence and affected scope before a response policy is applied.
Availability is part of the security contract
A response control exists only when the host protocol can reliably deny, preserve audit evidence, recover from failure, and pass compatibility tests. Roadmap controls should never be marketed as current enforcement.