It's not a question of if anyone holds the copyright, but a question of whether the models themselves and their outputs are derivative works of the data the models were trained on.

If they are derivatives, it may be possible that the licenses of that training data place additional restrictions on the generated code or forbid derviative works outright.

Speaking practically, for a random "FOSS" project that isn't very big, this is unlikely to ever come up. Courts would have to rule in favor of the training data owners, plus they would need to issue DMCA takedown requests and/or sue to have the content removed. That probably won't happen.

However, you should probably question any project where the developers don't understand the code well enough to write it themselves. I don't know if that's the case here, but that tends to be the case with vibecoded projects.