OpenAI has for some time not trained large language models like GPT on data obtained from API users. This was disclosed by the startup’s chief executive, Sam Altman, CNBC reports.
“Customers clearly want us not to train on their data, so we changed our plans: we won’t do that,” he said.
On March 1, OpenAI quietly updated its Terms of Service. However, the new policy applies only to customers who use the company’s API.
“We may use Content from Services other than our API,” according to the updated Terms of Service.
Altman confirmed the information:
“We have not been training on any API data, for some time.”
The company is likely still using texts entered into ChatGPT. Earlier, Amazon предупредила сотрудников not to share confidential information with the chatbot, fearing data leaks.
In May, Samsung banned dialogue AI systems on corporate devices and internal networks.
In April, reports said that company employees at least three times disclosed secret information about ChatGPT. The leak occurred soon after the ban on using the chatbot was lifted.