Report: Massive Amounts of Sensitive Data Being Shared with GenAI Tools

Harmonic Security’s analysis of 22.4 million prompts across six generative AI tools finds that 2.6% contained company-sensitive data, with most exposures occurring via ChatGPT Free; common leaked content included code (30%), legal discourse, M&A, and financial projections, and 17% involved personal/free accounts lacking enterprise oversight. The report warns that data collected by AI tools—sometimes stored in jurisdictions like China—may be used to train future models, potentially leading to latent breaches, though no PII or proprietary file content left customer environments in this study. It urges adoption of enterprise-grade tools with guardrails and continuous monitoring to reduce data leakage risk.