A policy that can’t be enforced is a liability document, not a security control. It tells you what employees shouldn’t do, not what they’re doing. Visibility follows the same pattern. Eighty-seven percent are confident in their count of AI applications, yet only two-thirds can see the specific data being uploaded to unsanctioned AI tools in real time. That’s like knowing which doors in the building are open without knowing what’s walking out of them.
The attack surface moved
Three trends are making this harder, and each disproportionately affects small teams. The first is mobile. Respondents estimate that 44% of employee AI interactions occur on mobile devices, yet only 20% report that their MTD or MDM tools provide granular visibility into data flowing between enterprise apps and AI assistants.
The second is SaaS-connected AI agents. On average, 36.7% of core SaaS tools, such as Slack, Zoom, and Salesforce, have been connected to third-party AI agents without formal IT review. Some 89% of security leaders are concerned about “shadow permissions,” in which an agent inherits an employee’s access to systems like OneDrive or Salesforce. An agent with a user’s permissions and no owner never appears in a traditional asset inventory.