AI is showing up in public health — but not always responsibly. A new commentary published in the American Journal of Public Health by physicians from the Uniformed Services University of Health Sciences and Tufts University School of Medicine warns that AI is being integrated into public health systems haphazardly, without adequate vetting or governance. The result? Real risks for patients — especially those from marginalized communities, including racial minorities, older adults, rural residents, and undocumented families.

The authors flag a range of concerns: chatbots trained on outdated or culturally insensitive data, generative AI that can produce false health information, deepfakes impersonating clinicians or public officials, and commercial apps that push uniform health recommendations while ignoring individual nuance. They also raise red flags about AI tools using location tracking and other commercial data, raising serious questions about consent and privacy.

Key Takeaways:

• AI tools in public health often prioritize technical capability and commercial interest over equity and accountability
• Biased training data can lead to misclassification and harm for already-underserved groups
• Risks vary by population — from children's developmental privacy to surveillance fears among undocumented individuals
• The authors call for mandatory equity-impact evaluations, national transparency standards, and independent third-party certification of AI tools

Why it matters: Without a coherent governance framework, AI could deepen existing health disparities rather than close them. The authors argue public health leaders must take an active role in shaping how these tools are designed and deployed — before the harms outpace the benefits.