AI Autonomy
Per-account, per-action-class settings that control whether the Iris AI assistant may apply a change unattended or must always ask you first. Autonomy only ever applies to safe, reversible action classes; every other change is confirm-gated regardless of these settings.
Two independent opt-ins are required before anything runs unattended: your plan must include the ai_autonomy entitlement, and an account owner must set a class to auto or canary here (the default for every class is confirm). Even then, an action is applied automatically only when a live traffic dry-run predicts it is safe — and it is monitored afterward and rolled back automatically if it hurts real traffic.
Human-only: these settings can only be changed by a signed-in account owner/admin. Iris itself can never modify them.
Modes
Configurable action classes
Only these reversible classes can be set to canary/auto: attack_ip_block, false_positive_relax, rate_limit_tune, firewall_rule_edit, traffic_policy_edit. All other classes always confirm.
Get Autonomy Settings
Returns the active account's setting for each configurable class (unset classes read back as confirm) plus whether the plan is entitled to autonomy.
Response (200 OK)
ai_autonomy entitlement. When false, unattended modes are refused.action_class, mode, canary_percent, canary_dwell_minutes.Update an Autonomy Setting
Upserts the mode for one action class. Requires an active account.
Request body
400.off | confirm | canary | auto. canary/auto require the ai_autonomy entitlement.60.Response (200 OK)
Echoes the stored setting (same shape as one classes entry).
Errors
action_class, invalid mode, or no active account.ai_autonomy entitlement.