The term is being used here to imply that the generated code is somehow bypassing the licensing requirements, which isn’t necessarily true, and certainly isn’t a substantiated claim.
You can read licensed code, learn from it, and then write your own code derived from that learning, without having committed a copyright violation.
You can also read licensed code, directly copy paste it into your codebase, and still not have committed a copyright violation, as long as you did so in a way that constituted fair use (which copy-pasting snippets certainly would).
There’s no copyright issue here at all, and rationally speaking there aren’t any legitimate misuse of open source concerns either. If these people were honest they’d just admit to feeling threatened by AI, but nobody would care about that, so they just try to manufacture some fake moral panic.