You can regression test your regression tests by running them against previous buggy versions of your program. If the tests do not detect the bug they were supposed to prevent, you have to fix them.
Isn't this a common bug fixing method? Write a test that fails due to the bug's behaviour, then update the code to make the test pass (ie. fix the bug)
It is, but I think the OC was suggesting that the old code can be kept around and used for automated regression meta-testing, which is an interesting idea.