Companies face data breach risk as employees feed sensitive info to ChatGPT

A 2025 LayerX report finds that 77% of employees regularly copy sensitive corporate data into generative AI chatbots such as ChatGPT, with an average of 14 AI prompts per worker per day and at least three containing protected information. The practice often occurs before firms have formal policies or technical controls, and responsibility for any resulting data loss remains with the employer. The European Union's AI Act now imposes stricter transparency and supervision duties on high‑risk AI systems, with potential fines up to €35 million, prompting a market for solutions that keep data on EU‑based servers and prevent it from being used to train AI models. Start‑up AIDAPT offers guardrails that automatically delete sensitive inputs, generate required DPIA documentation and log each step of AI‑generated responses to aid compliance.