I don't understand how code ownership will work. The application is "trained on ... source code from publicly available sources", but "code you write with its h

I don't understand how code ownership will work. The application is "trained on ... source code from publicly available sources", but "code you write with its help, belong to you". I get that most code it generates appears to be unique, but they admit that fragments of verbatim code occasionally appear. Unless the code they are using is all BSD-licensed, surely this breaches those licences?