As AI agents move from answering questions to executing actions, seeing AI activity isn't enough. Security teams need to control it.
AI requests now come from employee assistants, coding agents, SaaS copilots, and custom applications and cloud workloads that call models directly through SDKs. Each of these requesters connects to model providers on its own, with its own credentials, models, and settings.
Falcon AI Gateway, now generally available and included with CrowdStrike Falcon® Guardian™, extends AI visibility and enforcement beyond the CrowdStrike Falcon® sensor. It routes AI traffic through a centralized gateway to inspect prompts and responses, monitor usage, and enforce policy across managed and unmanaged devices.
One Control Point for Model Access, Usage, and Security #
Discovery shows where unsanctioned AI is in use. Falcon AI Gateway adds a complementary approach: Make the approved path the only path.
Falcon AI Gateway authenticates each request, routes it to the right provider and model, applies usage limits, and applies Falcon Guardian policy both before the request reaches the provider and again when the response comes back.
Teams block direct connections to model providers using the network controls they already run, and route approved applications and workloads through Falcon AI Gateway. As a result, the AI traffic reaching providers from those workloads has passed through Falcon Guardian policy on the way. This doesn't discover or redirect AI traffic automatically; it gives teams a deliberate control point for the workloads they choose to route through it.
This allows teams to:
- See : Traffic, models, tokens, and estimated spend in one place, with dashboards, request logs, and event data.
- Adapt : Applications send requests to one gateway endpoint with one gateway key, using the API formats they already speak. Provider credentials stay in the gateway, and administrators can switch between connected model providers, such as Anthropic, Amazon Bedrock, Microsoft Foundry, and Google Vertex AI, without reconfiguring every workload.
- Control : Rate limits and budgets on requests, tokens, or estimated spend, scoped by application, user, team, model, or provider. Application keys suit services and shared workloads; personal keys attribute usage to an individual.
- Secure : Falcon Guardian policy inspects prompts and responses routed through the gateway for risks such as prompt injection and sensitive data leakage.
Falcon AI Gateway and the Falcon sensor share one Falcon Guardian policy model, and each enforces where the policy sits. The sensor acts on the endpoint, including the local processes, files, and commands an agent runs. Falcon AI Gateway acts on traffic routed through it and can't stop local processes.
Because traffic flows through one point, AI Kill Switch in Falcon Guardian can cut off agents’ access to the model via the traffic routed through Falcon AI Gateway.
Falcon Guardian customers can find Falcon AI Gateway in the Falcon console under Falcon Experiences > AI Gateway. Developers connect an existing SDK or coding tool by changing the base URL to point at the gateway and using a gateway key.