Jesus, this just triggered another nightmare scenario I should have thought of earlier. People are going to have it write and/or comment a function whose purpose is non-trivial and not straight-forward. It will be not or imperfectly checked that the comment actually says what it should and the functionality matches. There will be no indication of what was written by AI, so the only option is to guess that both were written competently/with the same purpose, but when they don't actually match there's no way to know which is wrong except to check all other code that interacts with it and figure out what has been working right/expecting what was originally intended in ways that subtly break/have been working right depending on the incorrect implementation that existed.

This is absolutely something that already happens with fully human developers, but it seems likely to be much more frequent and not caught as soon with AI assistance.

This also seems like a failure mode that could go pathologically wrong on a regular basis for TDD types.