CrowdStrike Launches AI Agent Kill Switch and Acquires Fiddler and Dreadnode

On November 3, 2026, at Fal.Con Europe in Barcelona, CrowdStrike announced a stop button for AI agents. The AI Kill Switch has been available in Falcon Guardian since the announcement. The same day, the company said it was acquiring two companies: Fiddler AI and Dreadnode.

If your company already has, or is close to having, an agent answering customers, placing orders, or editing an inventory spreadsheet, the practical question is just one: who shuts it down if something goes wrong, and how quickly? That is the gap CrowdStrike set out to address.

What the kill switch actually does

An AI agent depends on a language model to decide and act. The kill switch cuts off that access. When an authorized security administrator triggers the feature, it instructs the connected AI "gateways" (the front door through which agents talk to models) to block access. That can be the Falcon AI Gateway, now generally available in Falcon Guardian, or a third-party gateway. The agent also loses access to the corporate network, and CrowdStrike says it is prevented from running again in the organization.

Every action triggered by an administrator is recorded in an audit log. For anyone who has to answer to an auditor, an insurer, or a customer asking "what happened?", that is documented proof.

George Kurtz, CEO and founder, said the button sits outside the model, at execution time, where the agent "can't argue with it, rewrite it, or wait it out." Chief product officer AJ Shipley said the decision is human and can be made with "surgical precision."

The limits you need to know about

The name promises more than the product delivers today, so it's worth reading carefully:

• It only acts at the AI front door. Today the cut happens at the gateway. Shutting down the agent directly on the computer where it runs is planned for a future release, with no published date.
• It cuts access, it doesn't delete the agent. According to CrowdStrike, as cited by TheStreet, the feature isolates the agent from the models that power its decisions rather than deleting its code.
• It depends on a person. Chief business officer Daniel Bernard described to CNBC a "circuit breaker, with a human making the decision." The feature does not act on its own.
• An agent outside the gateway is a blind spot. Yahoo Finance, in its own assessment and not CrowdStrike's, points out that it is unclear what happens with agents that bypass a compatible gateway.

The future vision is a single button coordinating agents across company computers, the cloud, user and agent identities, SaaS applications, and the browser. The company itself treats this as a vision, with no dates.

The two acquisitions and what they promise

CrowdStrike did not disclose prices or terms. The Wall Street Journal reported roughly $400 million for Fiddler AI and roughly $95 million for Dreadnode, $495 million in total. These are press figures, not confirmed by the company. The contracts are signed, but closing is expected in the fourth quarter of CrowdStrike's fiscal year 2027, which ends on January 31. Neither company is currently part of the products in operation.

Fiddler AI, which becomes Falcon Observe

Fiddler handles observability: tracking what each agent decided, which tools it used, and how it behaved. It evaluates agents before they go into operation, flags behavioral drift and hallucination (when the AI makes up an answer), and generates records useful for audits. In practice, it answers "why did the agent do that?" without depending on someone digging through logs across several systems.

Dreadnode, testing before launch

Dreadnode automates AI security testing. It runs more than 100 known attack methods, including jailbreak and prompt injection (tricks to make the AI disobey its rules), against models and agents before they go live, and ranks the results. For anyone about to put an agent in front of customers, it is the equivalent of a stress test before opening day. General availability is also expected in the fourth fiscal quarter of 2027.

What this changes for a mid-sized company

You don't need to buy CrowdStrike to take a lesson from the announcement. The lesson applies to any operation that uses agents: automation that acts on real systems needs limits, logging, and a way to stop. Performance, coverage, and pricing figures for these tools are not public, so they can't be compared with competitors. CrowdStrike claims to have the "industry's broadest set" of agent security, but that is the company's own statement, with no independent measurement.

In my analysis, the sensible path is a staged one, and it starts before any tool:

1. Inventory your agents. List which ones exist, including those someone on the team built on their own, and which systems each one can access: ERP, CRM, email, WhatsApp, invoice issuing.
2. Give each agent an owner. One person responsible for it and another authorized to shut it down, with names and contact details defined.
3. Limit what each agent can do. An agent that only checks inventory doesn't need permission to change prices.
4. Route everything through a single entry point. That is what makes it possible to cut access quickly, and it is exactly where CrowdStrike's kill switch works today.
5. Log and test. Keep a history of what the agent did and simulate failures before expanding its use.

This approach works by starting with a single painful process, validating the result, and only then extending it to others. The cost of a governance tool depends on the number of agents, the connected systems, and the level of monitoring, and can only be estimated by looking at the specific case.

CrowdStrike has given the next milestone on its timeline: the closing of both acquisitions and the arrival of Falcon Observe are expected by January 31, 2027.