LLMs loves to add dumb tests.
gpt-5.5 writes lots of great tests, in my limited experience, by looking at our 2 year corpus of manually written tests.
But it also loves to add tests for insignificant things, like "when we call the function, do we get the hard coded label" etc.