Sensitive Enterprise Data Uploads to AI Models Double in a Year

**Zscaler 2026 AI Threat Report:** Zscaler observed a 93% year-over-year increase in employees transferring sensitive enterprise data to AI/ML applications during 2025—about 18,033 TB total—resulting in hundreds of millions of DLP violations (over 410M tied to ChatGPT and 242M to Codium); exposed data includes PII, financial records, source code, and healthcare information, and the report recommends inventories of GenAI apps, disabling risky defaults, applying zero trust to model interactions, and enforcing inline inspection.