The other replies are correct that right now you need some level of human review, but it would be interesting to have a second AI review with a clean context. Maybe a security checklist, or a prompt telling it to check that the tests are covering the functionality appropriately.