Enterprise AI Risk Concentrated Among Small Group of Users

LayerX Security’s State of AI Usage Report 2026 analyzed enterprise AI interactions over the past year and found that AI activity and data exposure are concentrated among a small group of users and a few platforms.

The report shows nearly half of employees engaged with AI at least once during the year, while 18% used AI weekly. Usage is uneven: half of users had 12 AI conversations or fewer, and the top 5% of users generated at least 144 conversations.

Those top users also conducted deeper sessions, averaging 18 prompts per conversation compared with an average of two prompts for typical users. Nearly 30% of employees used multiple AI platforms, and the most active 5% interacted with six or more AI applications.

Platform activity was uneven. ChatGPT accounted for 36% of enterprise AI users and more than 55% of all AI conversations. Microsoft Copilot for M365 reached about 29% adoption and accounted for nearly a quarter of enterprise conversations. Much Gemini usage occurred through consumer accounts rather than enterprise-managed versions.

Personal accounts were a major channel for AI use. Almost half of enterprise AI conversations ran under personal identities, and over 14% of conversations tied to corporate identities used personal AI licenses. The report notes those patterns reduce visibility into retention policies, audit trails and governance controls.

Sensitive information appeared in more than 6% of enterprise AI conversations. Personal data appeared in 5.81% of conversations. Platform exposure rates varied: DeepSeek showed sensitive data in 12.63% of conversations, ChatGPT in 8.38%, and Copilot M365 in 3.65%.

The report flagged browser extensions and AI connectors as growing risk channels. About 15% of enterprise users run at least one AI browser extension; nearly 75% of those extensions request high or critical browser permissions and more than 16% have known vulnerabilities. AI connectors are being granted programmatic access to systems such as SharePoint, GitHub, Slack, Atlassian and Google Workspace, creating automated links between internal data and external AI services.

LayerX recommends that security teams focus monitoring and controls on the small group of high-risk users, expand visibility beyond approved enterprise AI tools to include extensions and connectors, block unmanaged personal account usage tied to corporate workflows, and adopt inline guardrails that can inspect prompts, uploads and responses in real time.

The report includes detailed metrics and use-case examples intended to help organizations map where AI exposure is occurring and which controls the report identifies for addressing those gaps.