Same experience here. The trap isn't the AI, it's that you stop thinking critically about what you're accepting. I've caught myself shipping patterns I wouldn't have chosen myself just because Claude made them convenient.
What helped: treat AI output like a junior's PR. Question the architecture choices, not just syntax. For React components, I ask "why hooks over context here" rather than just refactoring the generated code.
The paranoia is actually healthy. Keep it. Your code quality will drift if you don't actively push back on suggestions.