- Published on
CrowdStrike Launches AI Agent 'Kill Switch,' Plans to Acquire Fiddler AI
- Authors
- Name
- DeepHorus
CrowdStrike launched an AI agent kill switch and agreed to acquire AI startup Fiddler AI for about $400 million. The cybersecurity company announced the moves Tuesday at its Fal.Con Europe 2026 conference in Barcelona. The tool lets security administrators block unauthorized AI agents through a gateway, with direct endpoint termination planned later.
Businesses are increasingly deploying AI agents that can execute code, access sensitive information, and interact with business applications. Unlike chatbots, these agents can perform actions with limited human oversight. The same access that makes them useful can make their mistakes expensive, even without an attack.
The AI Kill Switch works through AI gateways, blocking an agent's access to the models powering its decisions. An authorized administrator triggers the kill switch, and the action is recorded in an audit trail. CrowdStrike plans to expand the feature to terminate agent processes on endpoints and coordinate restrictions across cloud environments.
Falcon Guardian can inspect agent prompts and apply policies to tools including Claude Code, OpenAI Codex, GitHub Copilot, Microsoft 365 Copilot, Cursor, and OpenCode. The risk is documented: Microsoft researchers found a vulnerability in Anthropic's Claude Code GitHub Action that could have exposed sensitive credentials. Anthropic later patched the flaw.
CrowdStrike also signed an agreement to acquire Fiddler AI, which specializes in monitoring AI models and explaining their behavior. The deal values Fiddler at roughly $400 million, according to The Wall Street Journal. CrowdStrike said the transaction would be paid primarily in cash, with some stock subject to vesting.
Fiddler's technology evaluates agents before deployment, traces their decisions and tool calls, and monitors behavior changes. The combined technology aims to give enterprises more control over AI agents that can access corporate systems and take actions without constant human supervision. Broader enforcement capabilities are still under development.
