It is a Tuesday morning. Half your coworkers are quietly using ChatGPT, Claude, or Gemini to draft emails, summarize meetings, and clean up reports. Most of them have never asked their manager whether they are allowed to.
That is the typical state of AI at work in 2026. According to a late-2025 report cited by IntuitionLabs, nearly 50% of enterprise employees use generative AI at work. According to research from Concentric AI, sensitive company data now makes up 34.8% of what employees paste into ChatGPT, up from 11% just two years earlier.
That is the problem. Not the AI itself. The pasting.
What gets people in trouble. According to IBM's 2025 Cost of a Data Breach Report, one in five companies has suffered a breach tied to "shadow AI", employees using AI tools the company never approved. The most common mistakes:
– Pasting client lists, customer emails, or phone numbers into a public chatbot
– Uploading internal financial documents or pricing models
– Sharing source code or proprietary product specs
– Putting confidential personnel matters or complaint investigations into a prompt
Samsung famously banned employees from public AI tools in 2023 after engineers pasted confidential semiconductor code into ChatGPT. Other major companies including Verizon, Amazon, and Accenture have followed with restrictions of their own.
What is usually fine. Most employers are not opposed to AI itself, they are opposed to confidential data leaving the building. Generally safe uses include:
– Polishing your own writing
– Brainstorming a problem
– Summarizing public material like news articles or industry reports
– Generating outlines, lists, or templates
– Learning a new skill or concept