Oversight modes
Per-agent autonomy levels that decide whether an action executes now or waits for a human.
category ▸ Governance
Oversight modes, in plain language.
Oversight modes make agent autonomy an explicit, configurable setting rather than an implicit property of the code. A mode decides, for each proposed action, whether the agent may execute it immediately or must wait for human approval — a dial from "suggest only" through to "autonomous within policy."
Crucially, oversight is layered on top of an action's own risk policy, and it can only make a decision stricter, never weaker. A high-risk action that must be approved stays approved-gated no matter how much autonomy an agent is granted; the action's floor always holds.
How Cortex implements it.
This term isn't abstract here — it maps to a real capability in the runtime. Here is exactly how Cortex enforces or relates to it.
Cortex offers five modes — suggest_only, draft, execute_low_risk, execute_with_review, and autonomous — resolved agent-specific first, then tenant default, then platform default (execute_low_risk). The action's high-risk / requires-approval gate is always a floor oversight can only tighten.
When a mode holds an action, the invocation lands pending_approval and an admin can approve, deny, or break-glass it with a recorded reason — every transition sealed in the ledger.
Keep building the vocabulary.
These terms sit next to this one in the governed-AI model — follow the thread to see how the controls connect.
Approval gate
A hold that pauses a high-risk action until a designated human approves it.
Break-glass
An audited override that lets an authorized human force a held action through, with a recorded reason.
Risk tier
A classification of how consequential an agent or action is, used to set the controls it needs.
Action Fabric
A lifecycle that makes every agent action a governed object: dry-run, propose, approve, execute, compensate.
See Oversight modes enforced, not just defined.
Book a walkthrough and watch the controls in this glossary return real verdicts, seal real evidence, and trace every fact back to its source.