As mentioned in the blog post, the statistics include IDE-related data, such as user actions, and basic information about the project, including frameworks and libraries (though limited to a predefined whitelist). Overall, this is quite similar to the data collection done by EAP for the products.
Source code is never sent under any circumstances.
There is enough of code on the internet already for big tech to train , what is probably missing is some scoring or labeling for that code so this AIs stop suggesting outdated stuff or bad stuff.
Source code is never sent under any circumstances.