# DeepSeek for healthcare: do no harm?

## TL;DR

This paper explores how biases in AI models like DeepSeek could negatively affect healthcare delivery and public trust.

## Contribution

The study introduces a novel analysis of 'pro-state' biases in AI and their specific implications for healthcare.

## Key findings

- AI models may contain biases that influence healthcare outcomes.
- Post-training methods and AI knowledge editing can introduce unknown risks in healthcare.
- Misuse of AI by powerful entities could harm public trust in healthcare technology.

## Abstract

Accessibility and cost remain barriers to the adoption of healthcare technology and will determine the impact of breakthroughs like generative AI. However, despite recent advancements in these areas, AI models may still contain biases and be prone to misuse by governments or other power structures with an interest in influencing public opinion. This report examines the potential effects of these “pro-state” biases on the delivery of healthcare. DeepSeek is used as a case study to illustrate the healthcare risks that may arise from unknown or biased post-training methods and other forms of AI knowledge editing.

The online version contains supplementary material available at 10.1007/s43681-025-00842-1.