It's a broad clause which relies on the fact that some types of computer software "use" may require permission of the copyright owners, depending on jurisdiction - in essence, I'd say that the validity of this restriction depends on how the specific law treats the incidental copies of the software created as it is being installed, executed, etc; this (unlike most parts of copyright principles aligned in international conventions) isn't universal globally.

My position is mostly based not on code but on text, as in natural language processing there is a similar but much older situation of models being trained on copyright-protected work, the interests of researchers and publishers obviously differ, and at least currently (laws do change) the legal position is that publishers' requirements can be (and are) ignored, as models can be trained on these texts without their permission and even after their explicit objections / cease and desist requests. And a BSD or GPL or some other license can't do anything more restrictive than the book "license" of "all rights reserved, we don't grant you any permissions".