core-concept · Security & governance · emerging · Reviewed
Observability & control
The combined telemetry and enforcement architecture for tracing agent behavior, evaluating policy, obtaining approval, constraining action, revoking authority, quarantining execution, and stopping systems.
Definition
The combined telemetry and enforcement architecture for tracing agent behavior, evaluating policy, obtaining approval, constraining action, revoking authority, quarantining execution, and stopping systems. Observability explains what happened. A control plane determines what is allowed to happen next.
Telemetry conventions are maturing; a unified agent control plane is not yet one settled standard.
Why it matters
Observability explains what happened. A control plane determines what is allowed to happen next.
The operating shift is from “Logs after execution” to “Evidence-aware intervention during execution.”
System anatomy
- Practice 1
- Trace model, tool, policy, approval, and outcome events
- Practice 2
- Keep deny, hold, revoke, and stop paths operational
- Practice 3
- Connect anomalies to accountable response
Important distinctions
- Logs after execution
- This concept moves the engineering system toward evidence-aware intervention during execution.
Implementation signals
- Trace model, tool, policy, approval, and outcome events
- Keep deny, hold, revoke, and stop paths operational
- Connect anomalies to accountable response
Failure modes
- Rich traces with no intervention authority
- A kill switch that is not continuously tested
Sources and further study
- NIST CAISI — Challenges in monitoring deployed AI systems
A current view of monitoring needs and limitations for deployed AI systems.
Use in this library: Official guidance. Guidance describes recommended practice; citation does not prove that a control is implemented or effective in a particular environment.
guidance · guidance · Published 2026-03-06 - OpenTelemetry — GenAI observability
Developing semantic conventions for model and agent telemetry across traces, metrics, and events.
Use in this library: Standard or protocol. A specification defines an interface or control pattern; conformance alone does not establish authorization, security, or fitness for a use case.
standard · standard · Published 2026-05-14 - OpenAI — How we monitor internal coding agents for misalignment
A current account of monitoring powerful coding agents that can act inside consequential development environments.
Use in this library: First-party case study. This first-party account documents one organization, product, or implementation context and should not be generalized without local evidence.
engineering · engineering · Published 2026-03-19