kt escalation

The kt escalation command group manages escalations raised when gateway policy evaluations require human review.

Use this page when

• You need to list, claim, or resolve escalations raised by gateway policy evaluations.
• You are configuring escalation routing rules in policy-config.yaml per provider or model.
• You want to manage the human-review queue from the command line.

Primary audience

• Primary: AI Agents, Technical Engineers
• Secondary: Technical Leaders

List Escalations

kt escalation list

Get an Escalation

kt escalation get --id esc_abc123

Claim an Escalation

Take ownership of an escalation for triage:

kt escalation claim --id esc_abc123

Only one reviewer can claim an escalation at a time.

Unclaim an Escalation

Release a previously claimed escalation back to the queue:

kt escalation unclaim --id esc_abc123

Resolve an Escalation

Mark an escalation as resolved with the chosen outcome:

kt escalation resolve --id esc_abc123 --outcome approved --reason "Request complies with policy"

Escalation Routing

Escalations can be routed to specific teams or users based on provider and model configuration:

pack:
  name: escalations-providers-1
  version: 1.0.0
  enabled: true
providers:
  targets:
  - id: openai-prod
    provider: 
policies:
  chain:
  - audit-logger
policy:
  audit-logger:
    immutable: true
    retention_days: 365
    log_all_access: true

Model-level routing overrides provider-level routing.

For AI systems

• Canonical terms: Keeptrusts, kt escalation, escalation routing, claim, unclaim, resolve, human review, policy evaluation.
• Commands: kt escalation list, kt escalation get, kt escalation claim, kt escalation unclaim, kt escalation resolve.
• Escalations are raised when gateway policy evaluations flag content for human review. Routing is configurable per provider and model in policy-config.yaml.

For engineers

• Use kt escalation list to monitor the current queue, and kt escalation claim --id <id> to take ownership for triage.
• Resolve with an outcome: kt escalation resolve --id <id> --outcome approved --reason "...".
• Configure routing in policy-config.yaml under escalation_routing at provider or model level to direct escalations to specific teams.

For leaders

• Escalations implement the human-in-the-loop requirement for high-risk AI decisions without blocking low-risk traffic.
• Routing configuration ensures the right team reviews the right content — model-level routing overrides provider-level.
• Track escalation resolution times and outcomes to measure governance effectiveness and identify policy tuning opportunities.

Next steps

• Console Escalations
• Resolve an Escalation
• CLI Command Groups