Enterprise AI use is growing faster than security teams can track it. Zscaler's 2026 AI Security Report, drawing on about 9,000 organizations, found transactions rose 83 percent and data transfers to AI applications surged 93 percent in 2025. A growing share happens outside formal approval, through shadow AI, built-in features and agents.
Enterprises are adding artificial intelligence tools faster than their security teams can track them, and a growing share of that activity happens outside formal approval channels. A new report from cloud security company Zscaler puts numbers behind the problem.
How fast is enterprise AI use growing?
Zscaler's ThreatLabz 2026 AI Security Report draws on nearly a trillion AI and machine learning transactions generated by about 9,000 organizations during 2025. Transactions across the platform climbed 83 percent year over year. The number of applications generating that traffic nearly quadrupled, topping 3,400.
The data side of the equation is growing even faster. Zscaler found that data transfers to AI and machine learning applications surged 93 percent in 2025. The company also counted 410 million data loss prevention policy violations tied to ChatGPT alone. Those violations included attempts to share Social Security numbers, source code and medical records.
AI is showing up in tools employees already trust
The traditional picture of enterprise AI is a sanctioned platform or a company-issued copilot that staff have been trained to use. That picture is incomplete. AI also arrives through features built into software the company already approved, through application programming interfaces that connect AI services to internal systems, and through autonomous agents that carry out tasks on their own.
Salesforce's Agentic Enterprise Index offers one measure of the shift. Between February 2025 and April 2026, Salesforce customers with agents in production more than doubled their agentic AI workforce.
Shadow AI adds another layer. An employee may install an unsanctioned tool to speed up a task, often without understanding how the tool handles data. Other times, an approved application rolls out an AI feature, and staff start using it without treating it as a new system that needs review. A Gartner survey of 302 cybersecurity leaders found that 69 percent said their organizations suspect or have evidence that employees are using prohibited public generative AI.
Why is shadow AI hard for security teams to track?
That is a problem for conventional tracking. Application inventories are built around software that is formally installed or purchased. A browser extension or a feature that switches on inside an existing application can slip past that list entirely.
Visibility has to reach the data and permissions
Knowing which AI tools exist is only part of the job. Security teams also need to know what information moves into those tools and what they can reach once connected. Prompts can contain proprietary code or customer details. A single file upload can expose an entire document that the employee never meant to share.
AI agents raise the stakes. Their reach often extends into whatever applications and permissions sit behind them. Knowing that a tool uses AI says little about whether it can pull from a customer database, write to a shared drive or pass data to other services.
What should security teams track besides AI tools?
Zscaler argues that effective discovery means tracking data flows and permissions alongside the tools themselves. The real risk, in the report's framing, lies in what AI can access and move, not simply in whether it is present.
The report also cautions against blanket restrictions. Blocking AI tends to push usage further underground rather than reduce it. A marketing team using an approved writing assistant on public content presents a different risk than an agent with access to financial records, and the rules should reflect that difference. Zscaler says controls should depend on the user, the application, data sensitivity, permission level and use case.
That approach ties back to Zero Trust principles that many security programs already follow: verify continuously, grant access based on context and never trust by default. The company says AI discovery needs to be an ongoing process rather than a one-time audit, because new capabilities keep appearing in existing software and agents keep taking on more autonomous roles.
For security leaders, the message is that governance policies are only as strong as the visibility behind them. Without a current view of which tools, applications and agents operate across the business, enforcing those policies becomes guesswork.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.