Sure you can. You could use an LLM to write a prototype or test cases, satisfy yourself that it works, then reimplement it by examining the generated code and deciding on your own implementation, and typing it out. You'd have to replace or throw out any LLM-generated test cases and other LLM-generated scaffolding, too, of course.

So you can use LLMs to write lots of code. It just can't be the code that you finally submit as the work you own and maintain. That might be a blunt instrument but it does seem likely to help ensure that compliant maintainers retain a (historically, at least) normal level of connection to their code regardless of what tools they use during development. I can imagine some developers who particularly care about the quality of some project choosing to impose this exact same discipline on themselves, even if for their dayjob or other personal projects they almost exclusively use LLM-generated code.

Maybe that's annoying, maybe it slows down development of the project, and it's fair for a maintainer to decide that's not how they want to work. But it's not the same as banning the use of LLMs to write code in connection with the project.

The comment I replied to said this:

> As AI bans go, this one doesn't seem too stringent if you disclose AI and actually check what it's doing.

I'm not arguing for or against the policy, but the policy as-written reads to me a complete and total ban on the submission of any and all generated text, which is not what that commenter suggested. I'm not sure you could possibly make it more stringent without getting into the territory of thought police.