Every day, millions of workers use Okta to gain access to applications and other services.
Soon, AI agents could require that same access. Okta’s CEO, Todd McKinnon, says his company is thinking about how to guarantee security as humans potentially pass off their digital responsibilities to AI agents.
He advocated for a kind of kill switch if, or when, things go rogue.
In an interview on The Verge’s “Decoder,” published on Monday, McKinnon described AI agents as a new class of digital workers — ones that can access systems, move data, and take actions across a company’s software stack. Businesses are increasingly experimenting with AI agents to automate workflows, build software, and even handle some physical tasks.
That kind of power needs strict parameters, McKinnon said.
“You need to have a system to keep track of them, define their role, define their permissions, and what they can connect to and what they can do,” he said, adding that companies will also need the ability to “pull the plug” if agents go rogue.
Okta’s proposal wouldn’t fully end AI agents. Instead, he advocates that the kill switch would minimize an agent’s access to sensitive data.
Harish Pari, the senior vice president of AI security at Okta, told Business Insider the risk is already emerging.
“Every organization is rolling out AI agents,” he said. “But for agents to really do their job, they need access to sensitive systems and data, thereby creating a new attack vector.”
While the upside is productivity, Okta says the risks are equally significant and require thoughtful boundaries, including the big red button. On March 15, the company published a press release titled “The blueprint for the secure agentic enterprise.” In it, Okta said AI-using firms should “be able to revoke access instantly across every system to contain risk.”
The framework also calls for real-time enforcement of data-sharing permissions, human approval for risky actions, and detailed audit logs that track every agent’s decision and access attempt.
Okta isn’t alone. In early 2024, California State Sen. Scott Wiener proposed an AI regulation bill that would require firms to build in a failsafe. The bill received support from AI backer Elon Musk before Gov. Gavin Newsom vetoed it later that year.
Still, McKinnon told “Decoder” it’s important for private companies to build their own failsafes.
“Stuff is going to go wrong, and there’s going to be issues, threats, and prompt injection,” he said, explaining that it’s important to stop agents from being able to access data in a crisis. “It’s almost like you would take a machine off the network.”
Read the original article on Business Insider
The post Okta’s CEO says all AI agents need a kill switch appeared first on Business Insider.
