That's funny in practice. But does not have to be a problem in principle.
There are ways to test your tests (in a way that does not lead to an infinite regress), and alternatively you can also prove your tests correct; or just use plain old code review.