How do I know when AI lab tools are wrong?

You use the AI tools now, and you have caught them being wrong, but you cannot always tell when. That is an unsettling place to work from, because in a lab a missed error does not stay in the lab; it follows a result out into someone's diagnosis or decision. So the fact that you are watching for it is exactly right. The tool can process and flag faster than you can, but it does not know when it is confidently mistaken. That is why it cannot be the final check, you are. Your value is not racing the machine; it is being the trained eye that knows what a result should look like and feels the wrongness when it does not. Keep that instinct sharp. So build the habit of pausing on anything that matters and asking whether it fits what you know to be true. Where it does not, trust yourself enough to look closer. The AI speeds the work; catching what it gets wrong is your work, and it is why the job still needs a human in the room.