I think it kind of flows into two trains of thought in the against category. First off, that some people are worried about copywrited, private stuff being included in the training data. I've not read up on copilot recently, so not sure if this is a reasonable thing to be worried about or not.

The other, is that people might be using Github to share stuff they've come up with other developers, but having an AI parse that information means that there's a disconnect between giver and receiver. It removes a chunk of the feedback loop being possible, which makes it so rather than it being a community of developers, it becomes something more akin to content creators and lurkers. That's not necessarily a bad thing, due to it opening up the sheer number of possible usages that end up using something. But it would minimize community feedback.

Many FOSS developers use copyleft licenses to ensure that only FOSS benefits from their code.

Copilot suggesting it for inclusion into proprietary codebases is then effectively whitewashing that GPL code.

Copilot also does not provide attribution, which is a legal requirement of tons of permissive licenses, including MIT and BSD.