CrowdStrike's new AI Kill Switch can stop a misbehaving AI agent, but only one part works today. Here is why that matters if agents touch your money or records.
CrowdStrike, a security company, announced a tool on Tuesday, Nov. 3, called AI Kill Switch. It lets an administrator stop an AI agent that is acting outside its job. An AI agent is software that carries out tasks on its own, such as paying bills or updating records. Only the first part of the tool is available now.
Every AI agent relies on a large language model, the AI engine that tells it what to do next. The agent sends the model what it sees, gets an instruction back, acts, and repeats. Companies route that traffic through a checkpoint called an AI gateway. When an administrator flips the switch, the gateway refuses the agent's requests. The agent keeps running but gets no instructions, so it stops acting.
A person has to trigger the switch, and each use is logged. The gateway can be CrowdStrike's own, included at no extra cost with its Falcon Guardian platform, or one a company already runs.
An agent that pays a vendor twice or rewrites a reconciliation record does it with a real employee or system login. Security tools see an authorized user doing an authorized task. In a January Cloud Security Alliance survey of 418 IT and security professionals, only 11% said they automatically block an agent that steps outside its job. Another 38% wait for a person to approve the action, and 24% just log it.
Microsoft takes a gradual approach. An agent starts with no permissions, gets a risk score as it works, and loses abilities in steps. It might lose the power to change records but keep the power to read them. Shutdown comes last. Cisco checks each request an agent sends and can block one bad request while the agent keeps working. CrowdStrike pulls the plug entirely.
CrowdStrike says it plans to shut agents down directly on the computers they run on, inside cloud systems and inside business applications. It also plans to revoke their permissions to company systems. The company said those features are still in development and may change, and it gave no release dates.
The company also said it is acquiring Dreadnode, which tests agents before they are deployed, and Fiddler, which monitors them afterward. PYMNTS, citing The Wall Street Journal, reported prices of about $95 million and $400 million. Both deals are expected to close by Jan. 31.
If your business lets AI agents touch payments or records, find out who can stop one quickly and how. Ask whether a stop happens automatically or waits on a person. Ask whether it cuts off the agent's access or only pauses its instructions.
If an AI agent can touch your payments or records, find out now who can stop it fast, and whether that stop is automatic.
Mastercard describes AI tools that could buy things for you, and says trust is the hard part: who gets to spend your money, and under what rules.
Read more: Mastercard outlines how AI assistants could shop and pay for youWorld models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasks
