Have you ever copied a product doc from Notion and pasted it into ChatGPT to summarize it for your team? It’s a common workflow, but it might not be as harmless as you think. Public Large Language Models (LLMs) like ChatGPT keep logs of user interactions, and some even reuse prompt data. This means that your internal documents could potentially be used as free training data for the world.
This might seem like a minor issue, but it can have serious consequences. For instance, Samsung learned the hard way when they lost $62 million due to an LLM security breach. The good news is that you don’t have to make the same mistake.
So, what can you do to protect your internal documents? One effective solution is to use a self-hosted LLM with open models that work well for summaries. You should also establish clear rules, such as avoiding client names, internal URLs, and proprietary code. Additionally, consider using context-aware redaction as a browser extension or proxy to further minimize the risk.
If you’re interested in learning more about LLM security risks and how to address them, I recommend checking out this article: https://tellian.io/2025/08/03/llm-security-in-2025-how-samsungs-62m-mistake-reveals-8-critical-risks-every-enterprise-must-address/. It provides a comprehensive list of risks and solutions to help you protect your internal documents.
By taking these precautions, you can ensure that your internal documents remain safe and secure.