I would say that what's most harmful is using code coverage as the primary measure of quality of your tests. It's that mindset that puts coders in a mode where they write tests which upon failure mean nothing significant (unless it finds a runtime error). It's a type of fallacy. Instead of considering if your tests verify your real-world requirements, you feel like your job is done because you have reached 100% line coverage.

It's like following someone's car and congratulating the driver that he drove correctly, without considering if he reached the correct destination.