An already failing test can stop before it reaches a new bug, giving an agent the same error and false reassurance. Failure masking means teams must recover evidence about the changed behaviour before accepting that the remaining failures are unrelated.