How do you validate AI-generated test code before committing it?

I've started using AI to generate first-pass test code. The problem: I've caught several subtle bugs in AI-generated tests: - A test that always passes because the assertion uses the wrong variable - A test that catches exceptions silently without failing - Fixtures that create data but assert on stale state I now review AI-generated tests more carefully than human-written code. But I'm wondering if there's a systematic approach: - Mutation testing on the generated tests? - A checklist of known AI test generation failure modes? - Run tests against known-bad code to verify they catch real bugs? What's your validation workflow?