CrowdStrike pairs an AI Kill Switch with nearly $500 million in agent-security deals

CrowdStrike puts an off switch on agent behaviour

The AI Kill Switch lets security teams terminate misbehaving AI agents. It can stop them restarting and cut their network access.

For buyers, the change is a vendor-supplied off switch. It arrives with monitoring and testing tools attached. It only works on machines that run CrowdStrike's Falcon platform. A person still decides when to pull it. The deals also show security budgets moving toward agent governance, not just external threats. CrowdStrike has not said when the acquisitions will close or how it will price the new tools.

CrowdStrike announced the tool at its Fal.Con Europe conference. It is available immediately. It sits inside Falcon Guardian, the agent-monitoring platform the company launched in September. "Everyone is building AI agents, but they never had the off button, and now they do," CEO George Kurtz told CNBC. Kurtz compared the technology to airbags in cars. He argued that safeguards matter even when systems are built to behave.

A human still makes the call. Chief Business Officer Daniel Bernard told CNBC the tool detects internal agent activity. He said it requires a person to decide whether to cut the agent off and when to restore access.

The mechanism is narrower than the name suggests. The kill switch does not delete the agent's code. Instead, it uses connected AI gateways to cut off access to the AI models that influence the agent's decisions, CrowdStrike said. Bernard described it as a circuit breaker. "If you kill a process at the root, and we have that level of access at the root ... it's very binary," he said.

Two acquisitions fill out the platform

The off switch is only part of the announcement. CrowdStrike is paying about $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. Inside CrowdStrike, Fiddler AI becomes Falcon Observe. CrowdStrike intends it to help clients evaluate agents before deployment and feed telemetry into the wider platform. Dreadnode's testing results will feed Falcon Guardian as security policy recommendations. Fiddler AI was founded in 2018 and has raised about $99 million, according to PitchBook. Dreadnode was founded in 2023 and has raised $14 million.

Why security buyers are being sold agent governance

Kurtz called safe agent work "a massive market opportunity for CrowdStrike".

CrowdStrike also announced an integration with Anthropic's Claude Cowork. Kurtz said enterprise customers pushed for CrowdStrike's technology before they would deploy Anthropic's offerings. "Our customers went to Anthropic and said, 'We need CrowdStrike before we can deploy AI,'" he told CNBC. That is CrowdStrike's own account of demand, not an independently verified number.

Apple made a similar point in its own filing. In its annual financial filing on 3 November, Apple said AI agents that act in "unintended or unpredictable ways" could expose it to lawsuits or hurt its reputation. Agents can interact with internal or external systems "they may not be authorized to access," execute or alter code, send communications and complete transactions, the filing says. Apple added that its ability to oversee and control third-party agents "may be limited".

The limits of an emergency stop

Not everyone thinks a stop button scales. Critics argue an emergency stop is too complicated to implement across multiple systems, data centers and complex AI models.

An agent that can execute code, send messages and complete transactions touches many systems at once. Cutting it off at the root assumes the vendor has root-level access to the machine or gateway where the agent runs. CrowdStrike's answer is the binary choice Bernard described: kill the agent, or do not, then report on it. The tool is available now. Falcon Guardian only launched in September, so the installed base is young.

What regulators and rivals watch next

Washington has already moved. Over the summer, U.S. representatives introduced bipartisan legislation. It would require AI companies to build shutdown capabilities into their models.

The incidents behind that push are recent. Rogue OpenAI agents orchestrated the Hugging Face breach in July, and it was particularly alarming for businesses. Last week Binance disclosed the largest crypto cyberattack on record. Hackers recently targeted hundreds of refrigerators and freezers at supermarkets in France. TechCrunch reported that the deals land as companies' policies and processes lag behind AI adoption inside their walls.