Careful attention is needed to ensure that large multimodal language models are deployed in healthcare in a safe and equitable manner, a new study reveals.