Kill Switches

Global operational controls to contain an AI incident in seconds — no redeploy, fully audited.

Kill Switches are emergency operational controls. They are not licensing and not feature flags — they exist to protect your environment during an incident, and take effect immediately with no redeploy or restart.

The switches#

Key	Effect
`agentExecutionDisabled`	Halt all AI agent execution.
`outboundMcpDisabled`	Block outbound MCP communication.
`externalLlmDisabled`	Block external LLM calls.
`promptExecutionDisabled`	Block prompt execution.
`policyEnforcementDisabled`	Disable policy enforcement.
`backgroundDiscoveryDisabled`	Pause background discovery.
`tokenIssuanceDisabled`	Stop issuing tokens.
`webhookDeliveryDisabled`	Pause webhook delivery.

Read state#

curl "https://your-company.authspoke.com/api/v1/support/kill-switches" -H "Authorization: Bearer $TOKEN"

Flip a switch#

Provider-only, audited as CRITICAL when disabling:

curl -X POST "https://your-company.authspoke.com/api/v1/support/kill-switches/externalLlmDisabled" \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{ "disabled": true }'

Operational properties#

Global — process-wide, affecting every tenant. Use deliberately.
Instant — backed by in-memory state, so a flip is visible to all request threads immediately.
Break-glass via JMX — switches are also exposed as a JMX MBean (com.authgateway:type=KillSwitches) so operators can flip them with no API call and no redeploy.
Audited — every change writes a KILL_SWITCH_CHANGED event.

Managing switches is restricted to AuthSpoke provider admins — see the Provider Console.