• 3 min read
Almost half (43%) of security pros say they will dedicate most of their behavioral-monitoring investments to AI agents in the next two years.

Francis Scialabba
Like first-time parents, a good percentage of security pros don’t understand what’s normal and what’s worrisome when it comes to their AI agent’s behavior.
According to a November report from behavioral AI security platform Abnormal AI, 43.6% of security professionals had a difficult time defining baseline, or “normal,” agent behavior. The findings are based on a survey of 525 cybersecurity professionals from organizations with more than 1,000 employees.
Where’s the disconnect? A majority (87%) of security teams claim behavioral baselining, or the process of identifying normal activities from users or systems, is “mostly or comprehensively operationalized” within their organization.
However, organizations still struggle with defining normal agent behavior because they don’t have a good grasp on their inventory of agents and governance around them, said Patricia Titus, field CISO at Abnormal AI.
“There aren’t a lot of companies that I’ve heard that have the governance structure in place,” Titus said. “They’re having to go back in time because the agents got basically out of the barn before we had the security controls in place.”
Another challenge facing security practitioners: current behavior detection tools often aren’t built to account for agent behavior. “My opinion is that they lean more toward human behavior,” Titus added.
Best behavior. Titus said it can be dicey for companies to not have a good understanding of what’s normal for their agents: “It’s really easy for companies to be socially engineered without having this baseline.”
She added that companies already struggle to identify “rogue” human behaviors, and that agents add another layer of complexity.
“Do we have the capability to identify AI that doesn’t belong in our environment?” Titus said. “To me, there is a major vulnerability and risk component to it that we haven’t even identified yet.”
Self-aware. Security teams understand the importance of behavioral detection. Almost half (43%) of the surveyed security professionals said they will dedicate most of their behavioral-monitoring investments to AI agents in the next two years.
In addition to these investments, Titus said, security professionals who want to establish a baseline of normal agent behavior should identify the full scope of agentic solutions in their organization and assign ownership over them.
“Who owns the responsibility for that agent?” Titus said. “Because they should baseline what’s normal, and help set up the guardrails and the policies.”