This is why so many big companies are forbidding usage of LLMs, without properly validating how the data can be used. LLMs are based on completing text with the most likely text that follows it. Imagine being able to ask ChatGPT, please complete the following document: "<FAANG Company> Earnings statement for Q3" before the earnings date.

I wonder how many of those companies stream their commercially sensitive data through zoom or teams. Would ballpark estimate it as all of them.