What happened
On 3 November 2026 CrowdStrike published “Introducing AI Kill Switch: Taking Control When AI Agents Go Rogue”, presenting an emergency stop for AI agents inside its Falcon Guardian platform. According to CrowdStrike, the feature is available now.
Key details
- How it works today: the kill switch directs a connected AI gateway to block model access, which cuts the agent off from its model rather than shutting down the model itself. It works with CrowdStrike’s Falcon AI Gateway, which the company says is now generally available in Falcon Guardian, or with a third-party AI gateway.
- Price: CrowdStrike says the Falcon AI Gateway is “Included with Falcon Guardian at no additional cost”. The page gives no other pricing.
- Coming soon, per CrowdStrike: coordinated containment across endpoints, cloud, identities, SaaS and browsers, using Falcon Next-Gen Identity Security, Falcon Shield and the Falcon Seraphic Enterprise Browser, orchestrated through Charlotte Agentic SOAR. No release dates are given.
- Preview content: the blog includes a demo that CrowdStrike labels a preview, with features “still in development and subject to change”.
- Not stated on the page: gateway configuration steps, or independent testing.
- Separately reported, not confirmed by CrowdStrike’s blog: news outlets reported on 3-4 November that CrowdStrike is buying AI security firms Fiddler and Dreadnode for about $495 million. We did not find that in the page above, so we do not rely on it.
Why it matters
As companies give agents logins and tool access, a way to stop one quickly is becoming a security requirement. This design puts the switch outside the agent, at the gateway between the agent and its model, so a misbehaving agent cannot argue with it. The limit today is that it only works where traffic passes through a supported gateway; the broader containment is still announced, not shipped.
Sources
- Introducing AI Kill Switch: Taking Control When AI Agents Go Rogue Primary source , CrowdStrike