The available control cuts model access

CrowdStrike made AI Kill Switch available to Falcon Guardian customers on November 3. An authorized administrator can use it to direct a connected AI gateway to deny an agent access to selected model providers. The issuer says the action is logged and can work with its own Falcon AI Gateway or a compatible third-party gateway.

Falcon AI Gateway is now generally available and included with Guardian, according to CrowdStrike. It centralizes model credentials, traffic inspection and policy enforcement for managed and unmanaged environments. The company says Guardian's agent-level hooks cover several prominent coding and workplace assistants, but it has not published a complete support matrix for every agent, gateway or model provider.

The endpoint promise is broader than today's release

CrowdStrike's announcement describes stopping an agent, preventing a relaunch and cutting network access. Its detailed availability material narrows the current termination point: enforcement begins at the gateway, while endpoint termination and coordinated containment across the wider Falcon platform are coming later. Gateway policy by agent type is also listed as forthcoming.

That boundary matters. Removing model access may interrupt an agent that depends on the governed connection, but it should not be treated as proof that every process has stopped. An agent using a local model, an alternate credential or an unmanaged route requires separate containment until CrowdStrike documents and ships the additional enforcement points.

Buyers need entitlement and recovery tests

CrowdStrike licenses Guardian by endpoint with an account-level monthly token allowance. The public licensing page explains the meter and optional additional capacity but gives no numeric base price. The announcement does not state country availability or an endpoint-containment release date.

A production test should verify administrator permissions, audit records, gateway coverage, blocked-agent reconnect behavior and the response when an agent changes models or credentials. CGH has not tested the control or confirmed its behavior with third-party gateways.