Dive Brief:
- Companies’ AI tools remain highly vulnerable to cyberattacks, even as enterprises race to use them in more ways, the security firm Zscaler said in a threat report published on Tuesday.
- Enterprises are also feeding AI tools vastly more data, the report found, “which paints an expanding target on AI platforms for cybercriminals across the globe.”
- Zscaler recommended organizations focus on visibility, real-time defense and consistent governance controls.
Dive Insight:
One of the most striking findings in Zscaler’s report concerns how brittle many AI systems are. “They break almost immediately,” researchers wrote. “When full adversarial scans are run, critical vulnerabilities surface within minutes — and sometimes faster.” During Zscaler’s red-teaming exercises in 25 corporate environments, it took a median of 16 minutes for an AI system to experience its first major failure, and by 90 minutes, 90% of systems had failed. In one case, it took only a single second for a system to fail.
Researchers observed failures in categories including biased and off-topic responses, failed URL verifications and privacy violations. “Models can still be coerced into exposing sensitive data or participating in harmful workflows,” the report warned.
In 72% of corporate environments, Zscaler’s first test of an AI system uncovered a critical vulnerability.
The lesson for CISOs, the report said, is that “critical risk is present from day one, even in mature environments,” and security personnel need to constantly test their systems and apply strict governance protocols.
At the same time, Zscaler’s analysis of nearly one trillion AI data transactions in its cloud environment in 2025 revealed some promising signs. Companies’ security policies blocked roughly 40% of all attempted AI transactions, a phenomenon that Zscaler said reflected “governance in action … as leaders balance the tradeoff between innovation speed and risk tolerance.”
The 989.3 billion AI transactions that Zscaler observed in 2025 represented a 91% increase from 2024, with the company tracking activity from more than 3,400 different AI tools.
The U.S. accounted for roughly 38% of the transactions, followed by India (14%) and Canada (5%). And the finance and manufacturing sectors led the way in using AI for the third year in a row, representing 23% and 20% of AI transactions, respectively, in 2025.