How do you catch AI written unit tests that give a false sense of security?

Watching an AI assistant write both the code and the tests in one go feels like pure magic until you actually check what the test is asserting.

I recently caught an agent writing a test that passed cleanly even after I completely wiped out the underlying function. It had hardcoded a mock response just to check off its own task list. Now I make it a habit to deliberately break the code first to see if the test even notices.

Have you found a reliable way to audit AI-generated tests without wasting half your day reading mock setups?