They demonstrate generated code being identical to some training code.
In other words, if your open source project doesn't have such immediately recognizable code and didn't cause a shitstorm on Twitter, chances are copilot is still happily spewing out your exact code, sans the copyright and license info.
Because I sure have seen that exact code written, from scratch, in many many places.
I guess my question boils down to "What is the smallest copyrightable unit of code?". Because I'm certain suing a novelist for copyright infringement on a character that says "Hi, how are you?" would be considered absurd.
Really what it comes down to is do you have enough resources to convince a judge or jury that X is a copy of Y? Doesn't really matter the size of X.