They will never be 100% accurate. They are like people. Even the smartest of people doesn't know things, has blind spots, has been trained incorrectly, etc. They are no different. We can trust them the way we can trust people. Perhaps eventually with a very high degree of confidence, but never with 100% blind trust.