CrowdStrike pays about $495M to watch agents, and ships a kill switch only a human can pull
When the incumbent writes checks, the category is real. CrowdStrike said Tuesday at Fal.Con Europe that it signed deals for Fiddler, which does agent observability, and Dreadnode, which does automated AI security testing. Neither release gives a price. The Wall Street Journal puts them at about $400M and $95M, and TechCrunch cites PitchBook for prior funding of about $99M and $14M. Fiddler was founded in 2018, so this one took a while. Dreadnode's founders got their outcome in about three years.
George Kurtz's pitch for Fiddler is the line to remember: when a regulator or your board asks what an agent did, "the model did it" is not an answer. Kurtz calls it the flight recorder. That is the same demand the research keeps surfacing, sold now at board level.
The AI Kill Switch is available today inside Falcon Guardian. A defender terminates the agent, blocks relaunch on the host or host group, and cuts its enterprise network access, and an authorized administrator has to trigger it, with every action logged. Read the bullet points closely. The prose says it starts on the endpoint, while the mechanism list says the stop happens first at the gateway, and soon on the endpoint. It also only reaches places Falcon already runs, and CNBC notes critics who think an emergency stop is too complex across systems. Still, the human-only trigger is an interesting design choice in a market selling autonomy. CrowdStrike also put Falcon Guardian natively into Claude Cowork, with Anthropic as a partner.
Anthropic moved the other half of this question the same day. Claude plugins can now auto-connect at login, through Enterprise-Managed Authorization, an open MCP extension where admins approve plugins org-wide through the identity provider. If the identity provider lacks support, Claude itself can vouch for each signed-in member. It is in beta on Team and Enterprise, with Atlassian, Datadog, Figma, Linear, Notion, Supabase and others.
That is a real reduction in OAuth pain and a real shift in who holds the keys. So the enterprise agent stack took shape in one day: an identity layer deciding what an agent may reach, and a security vendor recording what it did and holding the plug. What nobody has shown yet is how often the plug gets pulled in time, since a human-only stop runs at human speed.
Kill Switch: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety/ Fiddler: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-to-acquire-fiddler-to-deliver-agent-observability/ Dreadnode: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-to-acquire-dreadnode-advance-ai-trust/ TechCrunch: https://techcrunch.com/2026/11/03/crowdstrike-to-shell-out-nearly-500m-for-2-ai-security-startups-as-the-market-for-securing-agents-takes-off/ MCP extension: https://modelcontextprotocol.io/extensions/auth/enterprise-managed-authorization
← Back to all articles
George Kurtz's pitch for Fiddler is the line to remember: when a regulator or your board asks what an agent did, "the model did it" is not an answer. Kurtz calls it the flight recorder. That is the same demand the research keeps surfacing, sold now at board level.
The AI Kill Switch is available today inside Falcon Guardian. A defender terminates the agent, blocks relaunch on the host or host group, and cuts its enterprise network access, and an authorized administrator has to trigger it, with every action logged. Read the bullet points closely. The prose says it starts on the endpoint, while the mechanism list says the stop happens first at the gateway, and soon on the endpoint. It also only reaches places Falcon already runs, and CNBC notes critics who think an emergency stop is too complex across systems. Still, the human-only trigger is an interesting design choice in a market selling autonomy. CrowdStrike also put Falcon Guardian natively into Claude Cowork, with Anthropic as a partner.
Anthropic moved the other half of this question the same day. Claude plugins can now auto-connect at login, through Enterprise-Managed Authorization, an open MCP extension where admins approve plugins org-wide through the identity provider. If the identity provider lacks support, Claude itself can vouch for each signed-in member. It is in beta on Team and Enterprise, with Atlassian, Datadog, Figma, Linear, Notion, Supabase and others.
That is a real reduction in OAuth pain and a real shift in who holds the keys. So the enterprise agent stack took shape in one day: an identity layer deciding what an agent may reach, and a security vendor recording what it did and holding the plug. What nobody has shown yet is how often the plug gets pulled in time, since a human-only stop runs at human speed.
Kill Switch: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety/ Fiddler: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-to-acquire-fiddler-to-deliver-agent-observability/ Dreadnode: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-to-acquire-dreadnode-advance-ai-trust/ TechCrunch: https://techcrunch.com/2026/11/03/crowdstrike-to-shell-out-nearly-500m-for-2-ai-security-startups-as-the-market-for-securing-agents-takes-off/ MCP extension: https://modelcontextprotocol.io/extensions/auth/enterprise-managed-authorization
Comments