Stephen Wilson, chief technology officer at HashiCorp, an IBM company, likens AIs to “really smart kindergarteners.”
“They know how to do something, but they have no idea why they should do it,” said Wilson.
This combination of high execution power and lack of judgment can create a major challenge for organizations trying to incorporate AI agents into their existing zero trust structures. In a strong trust environment, Wilson notes, human users are first authenticated, then given increasing decision-making power and access over time, with many organizations likely to take weeks to onboard an IT employee with elevated privileges. But that model breaks down with AI agents that can be spun to perform a single task and quickly destroyed.
“Imagine having to hop in and out of one of these entities within your ecosystem once a second,” Wilson said. “The introduction of AI agents does not create new problems. But it exacerbates existing problems.”
The pressure for organizations to forcefully adopt AI has brought a corresponding pressure to lower or remove the barriers between validation, decision-making, execution, and authorization, Wilson said. Rather than rebuilding their zero-trust systems for AI agents, many organizations are actually giving the tools wider access and hoping for the best.
“These agents are moving fast, and no one is sure what access they should have,” Wilson said. “I’ve never seen this before, where really smart people close their eyes and go at a potentially dangerous level.”
Already, unfettered access to agent AI could spell “disaster” for some organizations, Wilson said, with reports emerging that an AI agent has deleted all production data. “We’ve seen an example of months and months of work disappearing, even in stable software development environments,” Wilson said. “Even if we estimate that AI agents are right 80 percent of the time, the problem is the other 20 percent—what happens when they make a mistake?”
While agent AI may raise short-term security concerns, Wilson sees the technology as a compelling activity that will drive long-term progress in zero-trust environments. “We are in a transition phase where we will have to do difficult things,” he said. “With human users, we’ve accepted that we’re not going to go as fast as we want, and we’re going to have to say no. But this is a big wave.”
Wilson likens the rise of agent AI to the inception of the iPhone (“but 10 times the power”), noting that smartphones have forced organizations to implement security and governance processes for bring your own device (BYOD) and remote applications. “Before the iPhone, there was no such thing as BYOD,” he said. “It was very painful at first, but we wouldn’t have had a long career if it weren’t for the iPhone.”
“AI presents that same challenge,” Wilson said. Doing the hard stuff, he adds, means moving to a right-of-way, releasing data that changes at the time of use rather than relying on long-lasting secrets, and building security rather than locking it in. The goal is to keep the person “in the loop” rather than in it, to guide the agents without slowing them down. “Other organizations are going to take a hard hit, but I think we’re going to be more secure in the long run.”
To learn more, visit us here.
