CrowdStrike unveiled an AI kill switch on November 3, 2026, a control that shuts down unauthorized AI agents in real time. The company also said it plans to acquire Fiddler, a startup that watches how AI agents make decisions. Both moves target the same worry: agents that take actions nobody approved.
The news came at the company’s Fal.Con Europe event in Barcelona. If your team lets AI agents write code or touch company data, this one is worth a few minutes.
What the AI Kill Switch Does
The kill switch lives inside Falcon Guardian, CrowdStrike’s tool for securing AI agents while they run. According to the company, it ends an unauthorized agent right away and blocks it from running again anywhere in the organization.
It works first at the AI Gateway, a checkpoint that routes AI traffic so prompts and answers can be inspected. CrowdStrike says the same shutdown will come to the endpoint, meaning the laptop or server where the agent actually runs, soon.
CrowdStrike president Michael Sentonas said agents often chain steps together, hit an obstacle and then take actions they were never authorized to take. That is the exact behavior the kill switch is built to stop.
Beyond the AI Kill Switch
The kill switch is one piece of a larger update. Guardian now combines three control points on a single sensor: the agent itself, the network traffic, and the device it runs on. Here is what CrowdStrike listed.
| Feature | What it does | Status |
|---|---|---|
| AI Kill Switch | Terminates unauthorized agents and blocks reruns | At the Gateway now, endpoint soon |
| AI Gateway | Inspects prompts and responses, enforces policy | Generally available |
| Agent-level hooks | Blocks malicious activity inside popular AI agents | Announced |
| Endpoint network inspection | Watches AI traffic leaving the device | Windows available, Mac in early access |
| AI supply chain protection | Finds malicious AI skills and MCP servers | Announced |
The hooks cover Claude Code, OpenAI Codex, GitHub Copilot, Microsoft 365 Copilot, OpenCode and Cursor. CrowdStrike also says it detects prompt injection and leaked personal data using its own models.
AI Agent Security Inside Claude Cowork
CrowdStrike is working with Anthropic so the Falcon sensor can run inside the Claude Cowork virtual machine. That gives security teams a view into what happens in that environment and ties it to activity across the rest of the company.
If your company is rolling out Claude, that matters. You can read our Claude overview for a refresher on what the assistant can do. CrowdStrike also says Guardian will extend to cloud workloads and frameworks such as AWS AgentCore.
Why CrowdStrike Plans to Acquire Fiddler
Stopping a bad agent is only half the job. Fiddler focuses on the other half: checking agents before they go live and explaining why they act the way they do afterward.
CrowdStrike says Fiddler will become Falcon Observe inside its platform, expected to be generally available in Q4 of its fiscal 2027. Financial terms were not shared in the sources we read, and the deal is still described as planned.
- Pre-deployment checks to catch agents that are misaligned before launch
- Tracing of decisions across the full agent hierarchy to spot drift
- Records that support audits and compliance reviews
- Coverage of both first-party and third-party agents
The pitch is simple. A security team can see not only what an agent did, but why. That pairing is a growing theme across the industry, as seen when agent security funding such as Pluto Security’s $42M raise.
What This Means for You
Most people will not buy this tool directly. Still, the announcement shows where AI security is heading, and a few habits are worth adopting now.
- List every AI agent your team uses, including coding assistants and browser tools.
- Decide what each agent may touch, then check logs for anything outside that line.
- Review the tips in our guide on Claude Code permissions before loosening any safety prompts.
- Ask vendors whether you can shut an agent off quickly if it misbehaves.
If you want a broader starting point, our piece on threat detection guide covers the basics.
What We Do Not Know Yet
These are vendor claims from a launch event. We have not seen independent tests of the kill switch, and the endpoint version is still listed as coming soon. Treat the speed and coverage claims as promises until customers report real results.
Sources
Primary source: CrowdStrike’s AI security announcement. Additional detail from CrowdStrike’s Fiddler acquisition post.
Conclusion
CrowdStrike is betting that companies will only trust AI agents if they can stop them instantly and explain what they did. The kill switch handles the stopping, and the planned Fiddler deal handles the explaining. Whether it works as promised will depend on real-world results, but the direction is clear: AI agents are becoming something security teams must manage like any other risky system.
