Companies are liable for errors in their AI systems just as they are for errors by employees—a rule that prevents AI deployment from being misused to evade liability for erroneous outputs.
A new benchmark enables identification of the exact point where medical AI models produce hallucinations and enables targeted countermeasures through trace-supervised fine-tuning.
