CrowdStrike opens Falcon AI Gateway to every Guardian customer

CrowdStrike opened Falcon AI Gateway to every customer of Falcon Guardian on 13 November. The gateway is one point that every AI request from an application can pass through. It carries CrowdStrike's security rules on the way in and on the way back.

The author is Vanessa Villa on the CrowdStrike blog. The post names four AI providers the gateway can route to: Anthropic, Amazon Bedrock, Microsoft Foundry and Google Vertex AI. It is in the Falcon console under Falcon Experiences, then AI Gateway. There is no separate price. The gateway is part of Falcon Guardian.

What the gateway does for a request

A request from an application reaches the gateway first. The gateway signs the request in. It sends the request to the right provider and the right model. It checks the request against Falcon Guardian policy before the model sees it. It then checks the model's answer against the same policy before the application sees it. The post names two risks the check looks for: prompt injection and leaks of sensitive data. The console carries dashboards, request logs and event data for every request it has served.

The controls a platform team gets

A platform team can set rate limits and spending caps. Each cap can be scoped to an application, a user, a team, a model or a provider. The team can issue application keys for a service. It can also issue personal keys for a staff member. Each request's cost and content is tied back to the thing that made it. The gateway also hooks into the Falcon AI Kill Switch. A security team can cut off an agent's access to any model from one place. The agent's code does not change.

What CrowdStrike says it has already run

CrowdStrike says it built the same gateway for its own staff. The post says the internal version has run for "several years." It names a number: "tens of thousands of CrowdStrike employees globally." Those are the company's own words. No outside party has measured the number or the policy's effect. The release is a vendor announcement, not a benchmark.

Where it fits for an AI engineering team

A team that already pays for Falcon Guardian can now set one policy for every AI request. That is new. A team that does not use Falcon Guardian gets no part of the release. The four providers named cover the three main US cloud model paths and Anthropic direct. A team routing through another provider has to wait. So does a team on its own fine-tuned endpoint or an open-weights model it hosts. Each of those keeps whatever gateway it uses today. The post does not name a date for more providers.

A team weighing this against a model-router gateway should remember the two do different jobs. A router sends a request to the cheapest model that can answer it. This gateway sends every request through one security check. A team can run both, one behind the other.

Source

Primary source: CrowdStrike blog: Falcon AI Gateway, by Vanessa Villa.