AI gave wrong output during project — what would you do?

During a project, AI started giving inconsistent results, which affected decisions.

Have you faced something similar? How did you handle it?