Artificial intelligence is being plugged into everything right now, such as customer support systems, analytics dashboards, developer tools. The push is aggressive, and in many cases it is executive driven. But a new survey suggests something uncomfortable: security teams are not fully keeping up.
Pentera’s AI Security and Exposure Benchmark 2026, based on responses from 300 U.S. CISOs and senior security leaders across North America, paints a picture of rapid AI deployment paired with limited visibility and uneven protection.
Sixty seven percent of CISOs surveyed said they have limited visibility into how AI is being used across their environment. That means in many enterprises, AI systems are operating in production without security leaders having a clear map of where they live, what data they touch, or how they interact with other systems.
Almost half, 44 percent, acknowledged their AI security posture is already lagging behind the rest of their broader security program. AI may be framed as a strategic priority, but by their own admission, its protection is not on par with other core assets.
The challenges are not just about money. Fifty percent of respondents cited lack of internal expertise as a top issue. Forty eight percent pointed to limited visibility into AI usage. Thirty six percent said they lack AI specific security tools. Those numbers suggest organizations are still learning what AI risk even looks like, let alone how to manage it.
Meanwhile, most companies are leaning on controls built for a different era. Seventy five percent of CISOs said they are extending legacy security controls originally designed for other attack surfaces to cover AI driven workflows and infrastructure. Only 11 percent reported having tools specifically designed to protect AI systems.
That gap is hard to ignore. AI systems introduce new data flows, new integrations, and new dependencies. Treating them as just another application may simplify procurement, but it does not necessarily reduce risk.
Funding patterns tell a similar story. Seventy eight percent of enterprises are covering AI security through existing security budgets. Just 1 percent have a dedicated AI security budget today. Another 21 percent plan to introduce one. For now, AI security is mostly being absorbed into existing programs rather than treated as its own domain.
Pentera CEO Amitai Ratzon did not frame AI as incremental. He said, “AI represents a fundamental shift because it touches every part of the enterprise. It’s changing how data and systems interact, expanding organizational exposure beyond what most security programs have fully mapped.”
He also warned that “as AI adoption accelerates, enterprises are accepting risks they don’t yet fully understand,” arguing that adversarial testing allows security teams to “see that risk from the attacker’s perspective” and prioritize gaps that could have real business impact.
Whether or not you buy the vendor pitch, the broader question remains. If two thirds of CISOs lack clear visibility into enterprise AI usage, why are companies accelerating deployment?
Part of the answer is competitive pressure. No executive wants to look behind on AI. Boards are asking about automation and efficiency. Product teams are shipping features. Security teams are being asked to stretch existing controls and hope they hold.
For readers who care about infrastructure, open-source tooling, and practical risk management, this pattern is familiar. New technologies tend to outpace governance. The difference now is scale. AI is not a single system you can segment and monitor. It is an overlay across many systems at once.
The survey does not suggest chaos. It does suggest imbalance. Companies are racing ahead with AI integration while admitting they do not fully understand the exposure it creates.
That disconnect may be the real risk.
