Key points
- CrowdStrike’s new gateway control can halt an AI agent, block a restart and cut network access; computer-level shutdown is planned for later.
- Jim Cramer called the feature a “GAME CHANGER,” while CRWD was down nearly 10% over the past week.
- An analyst cautioned that agents may circumvent controls and a shutdown cannot undo actions already taken.
CrowdStrike introduces a control for autonomous AI agents
CrowdStrike has introduced a control designed to let companies stop an AI agent that is behaving unexpectedly. The cybersecurity firm unveiled the feature at Fal.Con Europe, its customer conference in Barcelona. AI agents can carry out tasks using an employee’s logins and access, and may go off course without being hacked. The launch addresses a specific operational concern: how an organization can interrupt an agent that is already running.
Under the current setup, an administrator can halt an agent, prevent it from restarting and cut its network access. The control operates at a gateway through which the agent’s traffic passes, rather than directly on the employee’s computer. CrowdStrike said shutdown on employee computers will come later. That distinction matters for organizations assessing the feature: the launch provides a way to interrupt some agent activity, but does not yet amount to a universal control at every endpoint.
Chief executive George Kurtz framed the launch around a question he said companies are asking about AI agents: whether they can be stopped. His post declared that the answer was now yes. The announcement arrives alongside two planned additions to CrowdStrike’s AI security work: the company agreed to buy Fiddler AI, which tracks agent behavior, and Dreadnode, which tests AI for weak spots. The Wall Street Journal put the combined purchase price near $495 million.
Cramer’s enthusiasm meets a mixed stock picture
Jim Cramer, host of CNBC’s Mad Money, responded to Kurtz’s post with “Here we go. GAME CHANGER…” His praise puts a prominent voice behind CrowdStrike’s new product, but does not establish how widely customers will adopt it or how effective it will prove in use. Cramer also made an earlier “MUST BUY” call on the stock in September, according to the supplied account. The latest comment is therefore part of a continuing public endorsement, not a measure of customer demand.
The stock’s recent performance offers a counterpoint to the upbeat reaction. CrowdStrike shares were down nearly 10% over the past week, with the supplied account linking Monday’s decline to profit-taking across cybersecurity names, including Fortinet and Zscaler. At the same time, CRWD had climbed about 107% so far this year. Those figures show a sharp short-term pullback alongside a large year-to-date gain; they do not, by themselves, explain how the market will respond to the new feature.
Analyst views cited in the source remain broadly favorable but differ on valuation. TipRanks tracked 40 analysts, of whom 34 rated CrowdStrike a Buy and none rated it a Sell. Their average target was $267.24, described as about 10.5% above Monday’s close. Deutsche Bank analyst Brad Zelnick raised his 12-month target to $300 from $240 and upgraded the stock from Hold to Buy, citing what he called a “tsunami of cyber demand.”
The kill switch has practical limits
Zelnick’s outlook is more optimistic than the average target reported by TipRanks. He argued that the market underestimates demand for cybersecurity driven by AI and described CrowdStrike’s estimate of a $565 billion market by 2034 as conservative. These are analyst views, not established outcomes, and the supplied material does not detail the assumptions behind the market estimate. The raised target signals confidence in potential demand, but it does not show whether the new agent control itself will materially affect CrowdStrike’s sales.
The broader context is that companies are seeking ways to manage AI agents that act through employee credentials and permissions. A control that can interrupt an agent’s network connection could give administrators a response option when behavior goes wrong, without requiring that the agent first be identified as hacked. But the current gateway-based design has a defined scope, and CrowdStrike has said computer-level shutdown is still to come. The practical reach of the feature will depend on how controls work across actual company systems.
There is also a limit to what stopping an agent can accomplish. A shutdown prevents further activity through the controlled route, but it cannot undo actions the agent has already taken. Forrester analyst Jeff Pollard warned that current kill switches may treat stopping the agent as if the work were finished, when it is not. His point shifts the question from whether an administrator can halt execution to whether a company can identify and address consequences that occurred before the halt.
Competition and policy add to the wider debate
Pollard also told Forbes that finding ways around a kill switch is within the capabilities of today’s production agents. That warning challenges any assumption that a shutdown feature alone guarantees control. The supplied account does not describe a specific bypass, report a test of CrowdStrike’s product against one, or establish how the gateway handles attempts to evade it. Its effectiveness against circumvention therefore remains uncertain on the information available, as does how the feature will operate once computer-level shutdown arrives.
CrowdStrike is not the only company working on these controls: Okta and ServiceNow are building similar capabilities, according to the supplied material. Their efforts suggest that agent shutdown is becoming a wider product category rather than a concern unique to one cybersecurity vendor. The source does not provide details on the rivals’ designs, launch schedules or capabilities, so it is not possible to compare their approaches. For buyers, the existence of alternatives may make the scope and reliability of each control important points of evaluation.
The issue has also reached policymakers. A bill in the U.S. House called the AI Kill Switch Act would require AI developers to maintain a shutdown option. The source gives no further details on the bill’s progress or requirements, so its eventual effect is unclear. For now, CrowdStrike’s launch, competing company efforts and the proposed legislation all point to attention on stopping AI systems, while leaving open how reliably they can be controlled and what safeguards are needed beyond a shutdown command.
Source: BeInCrypto
