Stanford scholars respond to a federal RFC on evaluating AI-enabled medical devices, recommending policy interventions to help mitigate the harms of AI-powered chatbots used as therapists.

"If you’re following AI news, you’re probably getting whiplash. AI is a gold rush. AI is a bubble. AI is taking your job. AI can’t even read a clock. The 2026 AI Index from Stanford University’s Institute for Human-Centered Artificial Intelligence, AI’s annual report card, comes out today and cuts through some of that noise."

Permeation of artificial intelligence (AI) tools into health care tests traditional understandings of what patients should be told about their care. Despite the general importance of informed consent, decision support tools (eg, automatic electrocardiogram readers, rule-based risk classifiers, and UpToDate summaries) are not usually discussed with patients even though they affect treatment decisions. Should AI tools be treated similarly? The legal doctrine of informed consent requires disclosing information that is material to a reasonable patient’s decision to accept a health care service, and evidence suggests that many patients would think differently about care if they knew it was guided by AI. In recent surveys, 60% of US adults said they would be uncomfortable with their physician relying on AI,1 70% to 80% had low expectations AI would improve important aspects of their care,2 only one-third trusted health care systems to use AI responsibly,3 and 63% said it was very true that they would want to be notified about use of AI in their care.

In this testimony presented to the U.S. Senate Committee on Health, Education, Labor, and Pensions hearing titled “AI’s Potential to Support Patients, Workers, Children, and Families,” Russ Altman highlights opportunities for congressional support to make AI applications for patient care and drug discovery stronger, safer, and human-centered.

The annual report reveals a field hitting breakthrough capabilities while raising urgent questions about environmental costs, transparency, and who benefits from the technology.

Tabular medical datasets, like electronic health records (EHRs), biobanks, and structured clinical trial data, are rich sources of information with the potential to advance precision medicine and optimize patient care. However, real-world medical datasets have limited patient diversity and cannot simulate hypothetical outcomes, both of which are necessary for equitable and effective medical research. Fueled by recent advancements in machine learning, generative models offer a promising solution to these data limitations by generating enhanced synthetic data. This review highlights the potential of conditional generative models (CGMs) to create patient-specific synthetic data for a variety of precision medicine applications. We survey CGM approaches that tackle two medical applications: correcting for data representation biases and simulating digital health twins. We additionally explore how the surveyed methods handle modeling tabular medical data and briefly discuss evaluation criteria. Finally, we summarize the technical, medical, and ethical challenges that must be addressed before CGMs can be effectively and safely deployed in the medical field.