Research brief: An AI-based tool created by UCLA researchers had 99.8% accuracy in detecting potentially life-threatening errors, called realistic hallucinations, that occasionally come from virtual staining AI models.