Inputting company data into personal ChatGPT carries risks

The article addresses a practical dilemma for employees who use convenient AI tools like ChatGPT for work. The core issue is the default data usage policy: many cloud AI services train on user input, creating a leak risk when confidential company information is entered. This risk is present even for personal paid plans, unless the user or company takes specific action.

The article highlights two mitigation paths. Companies can contract enterprise plans with terms that exclude input data from training. Alternatively, some personal services allow users to change settings to prevent this. However, the article cautions against relying on personal accounts for work, as it creates 'shadow IT' — untracked usage that complicates investigations if a problem occurs.

As a forward-looking note, the article mentions local AI, or running models on machines under one's control, as a technology gaining attention to avoid data transmission risks. While historically inferior in performance to cloud AI, the gap has narrowed in the past one to two years, leading various companies to consider adoption. This suggests a shift toward more controlled AI deployment options for sensitive work.