CrowdStrike says the tool can now cut a misbehaving agent off from the AI model it relies on, with stopping agents on company computers coming “soon.”
Cybersecurity company CrowdStrike said Nov. 3 that it added a kill switch for AI agents to Falcon Guardian, its AI security product. CrowdStrike says a company’s security team can use the kill switch to stop an agent that does something it wasn’t approved to do. Companies need Falcon Guardian to use the kill switch.
The kill switch can only stop some agents
CrowdStrike says the kill switch first works by cutting an agent’s connection to its AI model through Falcon Guardian. It can’t stop agents that connect another way. CrowdStrike says stopping agents on the computers they run on will come later. Security Boulevard reported that, for now, the kill switch can only cut that connection. CrowdStrike hasn’t given a date for the rest.
CrowdStrike says the kill switch can also block restarts and cut network access
CrowdStrike says an administrator can also use the kill switch to keep a stopped agent from restarting. It says the administrator can cut the agent’s access to the company network. Security Boulevard reported that blocking restarts will be part of a later version for company computers.
Analysts say companies should test a kill switch before relying on it
Jeff Pollard, an analyst at research firm Forrester, told Forbes that stopping an agent doesn’t reverse its earlier actions. He said the agent may already have handed work to another agent or tool that keeps running. Pollard also said a kill switch is effective only if the agent can’t get around it. He said today’s agents can reason and adapt well enough to find ways around one.
Michele Abraham, an analyst at research firm IDC, told Forbes that companies should test a kill switch before they need it. She said companies should confirm that, once they use the kill switch, any agents the stopped agent handed tasks to are cut off as well. She also said they should test that they can restart the agent afterward.
Lawmakers and OpenAI previously proposed shutdown requirements
Rep. Ted Lieu (D-Calif.) and Rep. Nathaniel Moran (R-Texas) introduced the AI Kill Switch Act on July 23. Their bill would require the largest AI developers to keep the ability to slow down, suspend, or shut down their most advanced AI. It would cover companies that earn at least $500 million a year from such a system.
Under the bill, the Department of Homeland Security (DHS) could order a shutdown in an emergency. DHS could fine a developer up to $2 million a day for failing to keep a kill switch. DHS could fine a developer up to $20 million a day for ignoring a shutdown order. The bill is with the House Homeland Security Committee and hasn’t had a hearing or a vote.
Sen. John Kennedy (R-La.) introduced the AI Emergency Button Act in the Senate. It would require developers of advanced AI sold in the U.S. to build in a kill switch. Developers would decide when to use it. On Sept. 16, Kennedy asked the Senate to pass the bill through a fast-track process that any one senator can block. Sen. Rand Paul (R-Ky.) blocked it. Paul said Congress should understand the technology better before setting broad rules.
Both bills would require kill switches only from the companies that build AI. In an October recommendation to governments, OpenAI said businesses using AI agents should also be able to stop them.
