Making AI trustworthy: Can we overcome black-box hallucinations?