How could this have slipped through testing? Did their new version result in substantial enough code changes that the tests needed to be reimplemented but weren't? Of course, testing that kind of software/hardware interaction seems difficult.