Netskope CEO Sanjay Beri: Rogue Agents Make An AI ‘Kill Switch’ Mandatory
When it comes to AI agents that have the potential to go haywire, businesses can’t wait for regulators or AI platform providers to come to the rescue, Beri says during the latest episode of CRN’s Security or Else! podcast.
When it comes to AI agents that have the potential to go haywire, businesses can’t wait for regulators or AI platform providers to come to the rescue, according to Netskope co-founder and CEO Sanjay Beri.
For a special episode of the Security or Else! podcast series as part of CRN’s Cybersecurity Week 2026, Beri spoke about why companies now need their own AI kill switch and the key role that partners have to play in protecting against rogue agents.
The reality is, “companies have to realize that nobody is going to come and save them,” he said. “Nobody's going to regulate and legislate the fact that an agent can't go rogue or a model can't return poison content.”
That means businesses—through working with their vendors and partners—must take responsibility for securing their own AI agents against unexpected and unwanted behaviors that could lead to exposed data or compromised systems, according to Beri.
Beri—who led Netskope to one of the cybersecurity industry’s rare IPOs last year—also spoke about what is has been like to lead a newly public company through the massive growth in opportunities and challenges spurred by frontier AI over the past year.
For businesses adopting agents, a major shift is around the growing recognition that any agent, no matter how well-intentioned, can end up taking actions that are not intended or desired, Beri said.
“Agents don’t have a moral code,” he said.
The first step for partners and customers is to identify which tools are being used that are unsanctioned or risky, followed by introducing controls that can intervene when necessary with agents, he said.
“You implement your own visibility, your own controls, your own governance,” Beri said. “It has to be real time. It has to be in milliseconds.”
An AI kill switch, he said, could take several forms. These could include revoking an agent’s identity or stopping its activity at the endpoint or network level, according to Beri.
“Being able to, in a real-time way, detect rogue agents and get rid of them, stop them—that’s key,” he said. “And that’s kind of like a kill switch.”
Companies should apply a zero-trust approach to AI models, instead of simply assuming that their AI platform providers will prevent harmful behavior, Beri said.
For solution and service providers, the need for enhanced AI controls creates a huge opportunity, he said.
When it comes to customers, “they need help. They need advice. They need consulting,” he said. “And if you can be that trusted advisor who’s an expert, you have a golden role for the next decade in front of you.”
Check out the video here, and please like and subscribe if you enjoyed (available on YouTube, Spotify and Apple). Security or Else! features conversations with security experts and top executives about what’s hype and what’s real—and what channel partners truly need to focus on next in security and AI.