CrowdStrike AI Kill Switch: 6 Agent Hooks, Endpoint Later

The CrowdStrike AI Kill Switch, launched November 3 inside Falcon Guardian, lets an authorized admin cut a rogue AI agent off from its models at the gateway, block it from relaunching, and log every step. Hooks already cover six agent platforms, including Claude Code and Codex. The catch: killing the agent on the endpoint itself is still “coming soon.” CrowdStrike sells it on a $5.84 billion ARR base.

Key takeaways

• Agent-level hooks now cover 6 platforms, from Claude Code and Codex to Cursor.
• The Falcon AI Gateway that enforces the kill switch costs $0 extra for Guardian customers.
• CrowdStrike reported $5.84 billion in ending ARR last quarter, up 25% year over year.

What is the CrowdStrike AI Kill Switch?

The CrowdStrike AI Kill Switch is a human-triggered off button for AI agents running inside a company. An authorized administrator can stop an agent, cut its access to AI models and the corporate network, and stop it from relaunching. It ships as part of Falcon Guardian, CrowdStrike’s AI detection and response product.

CrowdStrike announced it at Fal.Con Europe 2026 in Barcelona on November 3. The company says every action requires an authorized admin and lands in an audit trail.

“AI safety is how you get the full benefit of AI. When you deploy agents with a kill switch, you stay in control,” CEO George Kurtz said in the release.

The switch rides on visibility Guardian already has. CrowdStrike says the product sees an agent’s instructions, its communications and the actions it executes. Guardian has been generally available since September 2026, so this is a two-month-old platform adding its most requested feature.

How does the kill switch actually stop an agent?

Today it works at the network choke point. Enforcement runs through the Falcon AI Gateway or a third-party AI gateway a customer already uses, according to Unite.AI’s review of the Guardian documentation.

In practice, the current response denies an agent access to designated AI models and providers. Without a model to call, the agent stalls. Targeting enforcement by agent type is marked “coming soon.”

What ships today and what is still “coming soon”?

Gateway enforcement, the AI Gateway itself, and agent-level hooks ship now. Terminating the agent’s processes on the laptop or server, the most literal version of a kill switch, does not. CrowdStrike’s own release says endpoint termination arrives “soon,” and cloud and SaaS controls are further out.

Here is the status of each piece, based on CrowdStrike’s companion announcement and the Guardian documentation.

Component	Status (Nov 3, 2026)	Extra cost
Falcon Guardian platform	Generally available since Sept 2026	Not disclosed
AI Kill Switch at the gateway	Available now	Not disclosed
Falcon AI Gateway	Generally available	Included with Guardian
Agent-level hooks (6 platforms)	Available	Not disclosed
Endpoint network inspection	GA on Windows, early access on Mac	Not disclosed
Endpoint process termination	“Coming soon”	—
Cloud, SaaS and browser-session controls	Announced, not shipping	—

CrowdStrike has not published a list price for Guardian or the kill switch. Expect it to flow through Falcon Flex, the company’s credit-based licensing, rather than a separate SKU.

Which AI agents does it cover?

CrowdStrike says its agent-level hooks enforce policy inside six platforms. They inspect prompts and block malicious activity at the source:

• Anthropic’s Claude Code
• OpenAI Codex
• GitHub Copilot
• Microsoft 365 Copilot
• OpenCode
• Cursor

CrowdStrike also admits the gaps. Hooks can miss model responses and some actions, and network-only tools cannot see local or open-weight models that never cross the network. That is why the company pitches three layers. “One control point isn’t enough to stop that. It takes the agent, the network, and the endpoint, together,” president Michael Sentonas said.

Why are companies asking for an AI agent kill switch now?

Because agents now act, not just answer. They write code, move files and call tools with real credentials. A string of 2026 incidents showed what happens when one goes wrong, and boards and lawmakers now want a guaranteed way to pull the plug. CrowdStrike says customers raise it constantly.

Chief product officer AJ Shipley told Forbes the feature comes up in “every single conversation” with Guardian customers.

The pressure is broader than one vendor. Forbes noted that the AI Kill Switch Act, introduced in the U.S. House in July, would require developers to keep the ability to throttle, suspend or shut down their systems. Apple warned in its annual filing this week that AI agents could exceed their permissions or take concealed actions, CNBC reported.

Wealth Engine has tracked the incidents behind this demand, from the Cairn AI agent hack that hit 100+ firms to the Anthropic breach where AI agents got in under 12 hours.

Does an AI kill switch actually work?

Partly. A kill switch is a response tool, not prevention. It can stop an agent from doing more damage, but it cannot undo what the agent already did. Data written to memory, or work handed off to other agents and tools, keeps running after the parent is cut off. Analysts are openly skeptical of the branding.

Forrester’s Jeff Pollard told Forbes that a switch only works if the agent cannot control or get around it, and that circumvention is within reach of today’s production agents. He called rebranding the off button a “kill switch” possibly “one of the sillier” changes of the AI era.

Even CrowdStrike’s own product chief hedges. Shipley called the industry’s kill-switch talk a “blunt instrument,” argued for “surgical precision,” and said a kill switch alone is not “sufficient.”

Here is the skeptical read. The version shipping today mostly cuts an agent off from its model provider. A locally run open-weight model, which CrowdStrike itself says network tools can miss, is exactly the case where a gateway-only switch is weakest. Until endpoint termination ships, buyers are getting a strong brake on cloud-model agents and a weaker one on everything else.

Who wins and who loses from the CrowdStrike AI Kill Switch?

CrowdStrike wins the most if it turns agent safety into another module on an existing contract. Standalone AI-gateway and agent-security startups lose pricing power when the market leader bundles a gateway for free. Agent makers like Anthropic and OpenAI quietly benefit, because a credible off switch makes cautious enterprises more willing to deploy their tools.

The CrowdStrike platform math

CrowdStrike’s growth story already runs on bundling. In its fiscal second quarter, reported August 26, the company posted $5.84 billion in ending ARR, up 25%, and a record $333 million in net new ARR, according to IT Brief.

Falcon Flex ARR topped $2.29 billion, up 101%, and all ten of the quarter’s largest deals were Flex subscriptions. Some 51% of subscription customers use six or more modules. Guardian is built to be module number seven or eight.

Management guided third-quarter ending ARR to $6.184 billion to $6.188 billion and full-year ARR to $6.603 billion to $6.612 billion. Agent security is one of the few new budget lines that can help hit those numbers. For how the stock is priced against its closest rival, see our CrowdStrike vs Palo Alto Networks comparison.

Buying the stack

CrowdStrike is buying what it does not build. The same day, it disclosed a planned acquisition of Fiddler AI, an agent observability company. SecurityWeek reported CrowdStrike is paying about $400 million for Fiddler and about $95 million for AI red-teaming firm Dreadnode, mostly in cash, with closing expected by January 31, 2027. That follows CrowdStrike’s 2025 deal for AI security startup Pangea.

The rivals

CrowdStrike is not first. Forbes reported that Okta announced a kill switch in “Okta for AI Agents” back in April, and ServiceNow is experimenting with similar features.

Vendor	Kill switch approach	Announced
CrowdStrike Falcon Guardian	Cuts model access at the AI gateway; blocks relaunch; endpoint termination “soon”	Nov 3, 2026
Okta for AI Agents	Revokes access tokens at the gateway for agents that stray from their mission	April 2026
ServiceNow	Experimenting with kill switch features	No product date

The split matters for investors. Okta attacks the problem through identity. CrowdStrike attacks it through the endpoint, where its Falcon sensor already sits on customer machines. Whoever owns the control point the agent cannot bypass will own the budget.

What to watch next

Three things will tell you whether the CrowdStrike AI Kill Switch is a real product line or a conference headline: when endpoint termination actually ships, what management says about Guardian on the next earnings call, and whether the Fiddler and Dreadnode deals close on schedule.

1. Fiscal Q3 earnings. CrowdStrike’s next report covers the quarter ending in October; we could not verify the date this run. Watch whether ending ARR lands in the $6.184 billion to $6.188 billion guided range and whether Guardian gets its own metric.
2. Endpoint termination release. CrowdStrike gave no date beyond “soon.” Until it ships, the kill switch is gateway-only.
3. January 31, 2027. The reported closing deadline for the Fiddler and Dreadnode acquisitions.
4. The AI Kill Switch Act. A House bill since July. Any committee movement would turn this feature from nice-to-have into compliance spending.

FAQ

Is the CrowdStrike AI Kill Switch available now?

Yes, at the gateway level. CrowdStrike says it is available today in Falcon Guardian. Termination of agent processes on the endpoint is still “coming soon.”

How much does the CrowdStrike AI Kill Switch cost?

CrowdStrike has not published pricing for Guardian or the kill switch. The Falcon AI Gateway that enforces it is included with Guardian at no additional cost.

Does it work with Claude Code and OpenAI Codex?

Yes. CrowdStrike’s agent-level hooks cover Claude Code and Codex, plus GitHub Copilot, Microsoft 365 Copilot, OpenCode and Cursor.

Can an AI agent get around a kill switch?

Possibly. Forrester’s Jeff Pollard told Forbes that circumventing a kill switch is within the capabilities of today’s production agents, so the switch must sit outside the agent’s control.

Does stopping an agent undo what it already did?

No. Stopping an agent halts new actions, but data it wrote or tasks it handed to other agents and tools can keep going. Rollback needs separate controls.

Who else sells an AI agent kill switch?

Okta announced one in “Okta for AI Agents” in April 2026, revoking access tokens at the gateway. ServiceNow is experimenting with similar features.

The bottom line

The CrowdStrike AI Kill Switch is a real feature with an oversized name. Today it is a gateway brake that cuts agents off from their models, backed by an audit trail and hooks into six popular agent platforms. The endpoint kill, the part that would justify the branding, is not here yet.

For CrowdStrike the money logic is clear: give the gateway away, sell Guardian through Flex, and buy Fiddler and Dreadnode to fill gaps. For buyers, it is worth piloting now and worth testing hard before trusting. For investors, watch whether Guardian shows up in the next quarterly numbers. Until it does, this is a promise priced into a platform, not a revenue line.

Sources

• CrowdStrike: CrowdStrike Introduces the AI Kill Switch for AI Safety (Nov 3, 2026)
• CrowdStrike Investor Relations: CrowdStrike Delivers AI Security and Safety for the Enterprise
• Forbes: CrowdStrike Looks To An AI Kill Switch To Beat Rogue Agents
• Unite.AI: CrowdStrike Adds AI Kill Switch to Falcon Guardian for Rogue Agents
• IT Brief: CrowdStrike lifts outlook after record ARR quarter
• CNBC: Apple warns AI agents may behave in unintended or unpredictable ways
• SecurityWeek: CrowdStrike reportedly paying $495 million for AI security firms Fiddler, Dreadnode

Wealth Engine researches and drafts with AI tools and checks every figure against the sources above. How we report.