It’s not as if I haven’t warned you here often enough. And now one of the first major AI data protection scandals has come to light.
The key findings of the “2025 Data Risk Report” show* that Copilot accesses almost three million confidential records per company on average, which corresponds to roughly 55% of all externally shared files (for the TL;DR crowd, here’s a shortcut to the summary of this article as a promptcast)
Industries such as financial services and healthcare are hit particularly hard, with the share of privileged information at around 70%.
The situation is made worse by data management practices that were already poor, including the mass storage of duplicate, outdated and orphaned records, and excessive data sharing without adequate restrictions. The report warns in no uncertain terms that organisations without stronger governance will struggle to effectively protect their intellectual property, financial information and personal data from an external AI.
This isn’t just embarrassing for those affected (never mind the loss of business secrets). Since 2 August 2025, it’s also a clear breach of the law.
What to do
The quickest fix is to get to grips with this useful rulebook right away: The purpose of the “General Purpose AI Code Of Practice”, and to set up the AI governance you need, above all to get “shadow AI” under control. Shadow AI refers to employees using AI applications and tools such as ChatGPT, Gemini, etc. without official approval or oversight from the IT department or management.
Summary of this article as a promptcast
*Original source
As data sharing risks escalate, Copilot found to be accessing millions of confidential business recordsNew report shows just how problematic AI in business can beWayne Williams · TechRadar
🚀 Service: Book your own AI expert for free 😉
