CrowdStrike Unveils AI Kill Switch and Agent Governance Framework

Cybersecurity firm CrowdStrike announced its AI Kill Switch at the Fal.Con Europe 2026 conference in Barcelona on November 3. The company noted that as AI agents gain system-level access across endpoints, clouds, and SaaS platforms, they risk executing unauthorized actions without human oversight. According to press releases and media reports, the kill switch currently operates at the AI Gateway layer between agents and AI models, with endpoint-level blocking slated for a future release. Rather than deleting agent code, the switch cuts off access to the AI models that drive decision-making, while Falcon Guardian enforces policies to prevent restarts and isolate network access. Actions require authorization from administrators and are logged in audit trails. Additionally, CrowdStrike integrated sensors into Anthropic's Claude Cowork virtual machines and announced plans to acquire Fiddler and Dreadnode to expand pre-deployment observability and automated security testing.

Key takeaways

• CrowdStrike launched an AI Kill Switch triggered by authorized administrators to cut off model connections via an AI Gateway, with endpoint blocking coming later.
• The defense includes blocking agent restarts and network access, using hooks in tools like Claude Code and Copilot to inspect prompts and intercept threats.
• CrowdStrike partnered with Anthropic to deploy sensors in Claude Cowork and signed agreements to acquire Fiddler and Dreadnode for observability and testing.