CrowdStrike adds an AI kill switch and buys two agent-security firms

CrowdStrike puts a kill switch in Falcon Guardian

CrowdStrike built the kill switch into Falcon Guardian. CrowdStrike launched that platform in September to secure AI agents.

Critics argue an emergency stop is hard to implement across multiple systems, data centers and complex models. The tool is designed to stop internal agents that go rogue, block them from relaunching and cut off their enterprise network access.

It does not delete the agent's code. Instead, it uses connected AI gateways to cut off access to the AI models that influence an agent's decisions, the company said. A human has to decide to cut the agent off and to reinstate access. Chief Business Officer Daniel Bernard told CNBC the process is "very binary": "You either kill the AI and the agent, or you don't, and then you report on that".

CrowdStrike also announced an integration with Anthropic's Claude Cowork.

The bet on seeing and testing agents

The two deals cover the parts a kill switch cannot reach.

CrowdStrike will pay around $400 million for Fiddler AI and about $95 million for Dreadnode, according to The Wall Street Journal, as cited by TechCrunch. Fiddler AI makes observability software for monitoring AI agents inside enterprises. Dreadnode automates security testing aimed at finding flaws in AI models, apps and agents before they go live. Fiddler AI will become Falcon Observe and feed telemetry into the Falcon platform. CrowdStrike will fold Dreadnode into Falcon Guardian.

Fiddler AI was founded in 2018 and has raised about $99 million, according to PitchBook. Dreadnode, founded in 2023, has raised $14 million.

CrowdStrike and its rivals are buying cybersecurity startups fast as demand for AI security tools grows.

Apple and Google widen the risk case

Apple added agent risk to its annual filing on 3 November 2026.

The company said AI agents that can act in "unintended or unpredictable ways" could expose it to lawsuits or hurt its reputation. Apple wrote that agents can take actions autonomously. They may interact with internal or external systems "that in some cases they may not be authorized to access". They can also execute or alter code, send communications and complete transactions. The filing also warns that Apple's "ability to oversee and control third-party artificial intelligence agents and other technologies may be limited".

Apple does not currently have an agentic AI. On Monday's earnings call, new chief executive John Ternus called agentic capabilities "promising tools for users". He said Apple wants to make sure users know what they are giving access to.

Google Threat Intelligence Group and Mandiant reported on Monday that LLMjacking attacks produced fraudulent cloud charges above $100,000 in a single day at some organizations. Cumulative charges reached $1 million at some organizations before access was revoked. Attackers get in through stolen cloud identity tokens, exposed API keys and misconfigured environments. The report named UNC6993. It said the group breached a cloud instance through a misconfigured service account key and used it to reach an LLM API.

SentinelOne launched Singularity Agentic Detection and Response on 29 October 2026. It can contain a single AI agent without stopping the wider workflow.

Governance is the other control layer

Snowflake pushed a different control layer at its Expedition 2026 conference on 3 November 2026. The target is the definitions and policies an agent works from.

Business Ontology sits inside Snowflake Horizon Catalog and serves only approved business definitions to agents, so answers stay consistent and traceable. Snowflake marked it for public preview soon. Cortex AI Gateway applies policies to agent connections across clouds, frameworks and AI providers, with controls for tool-call authorization, activity records and spending limits. Cortex AI Gateway Tools is in public preview, while Cortex Sandboxes, Agent Abuse Detection, Agent Trace Intelligence and Dynamic Model Routing are in private preview.

The pitch rests on a familiar complaint. In an MIT Technology Review Insights report sponsored by Snowflake, 55% of executives cited a lack of business context as the main cause of low-quality agent insights. Another 45% cited fragmented or inaccessible data.

Snowflake also said CoCo Cloud Agents and CoCo Automations are generally available for recurring data-engineering work, including checking pipeline health and investigating failed jobs. Those agents can propose fixes for engineers to review.

What is still missing

CrowdStrike gave no closing dates for either deal.

SentinelOne's ADR is in early access.

Two Partnership on AI researchers made the case for observability in a 29 October opinion piece. Monitoring requires observability, they wrote. Madhulika Srikumar and Claire Leibowicz argued that means knowing what an agent did and why it did it.