This isn’t a danger, it’s a benefit: your tests of one unit (whatever you consider a “unit” to be) are isolated from the behaviour of the rest of the system. For example, if you break one part of your implementation, you get one failing unit test, not a thousand.

Checking these assumptions is what integration and acceptance tests are for.