This voice is automatically generated. Please let us know if you have any feedback.
Diving overview:
- Even as companies race to use AI tools in more ways, their AI tools remain highly vulnerable to cyberattacks, says security firm Zscaler. mentioned in the threat report Published on Tuesday.
- The report also found that companies are feeding massive amounts of data to AI tools, “expanding the targeting of cybercriminals around the world on AI platforms.”
- Zscaler recommends that organizations focus on visibility, real-time protection, and consistent governance controls.
Dive Insight:
One of the most striking findings in Zscaler’s report concerns how vulnerable many AI systems are. “They break down almost immediately,” the researchers wrote. “When you run a full adversarial scan, critical vulnerabilities surface within minutes, sometimes sooner.” When Zscaler conducted red team exercises in 25 enterprise environments, it took a median of 16 minutes for AI systems to experience their first major failure, and 90% of systems failed by 90 minutes. In one case, it took just one second for the system to fail.
Researchers observed failures in categories such as biased or off-topic responses, URL validation failures, and privacy violations. “Models may still be forced to disclose sensitive data or participate in harmful workflows,” the report warns.
Initial testing of Zscaler’s AI system found that in 72% of enterprise environments; critical vulnerability.
The takeaway for CISOs, according to the report, is that “significant risks are present from day one, even in mature environments,” and that security professionals must constantly test and rigorously enforce systems. governance protocol.
At the same time, Zscaler’s analysis of nearly 1 trillion AI data transactions in cloud environments in 2025 reveals some promising signs. Corporate security policies blocked about 40% of all attempted AI transactions, a phenomenon that reflects “governance at work as leaders balance the trade-off between speed of innovation and risk tolerance,” Zuscaler said.
Zscaler observed 989.3 billion AI transactions in 2025, a 91% increase from 2024, and the company tracks activity from more than 3,400 different AI tools.
The US accounted for approximately 38% of transactions, followed by India (14%) and Canada (5%). Additionally, the financial and manufacturing sectors have led the use of AI for the third year in a row and will account for 23% and 20% of AI transactions, respectively, in 2025.
