Predicting pandemics, predicting people: the ethical limits of AI-powered public health surveillance

Artificial intelligence (AI) is transforming public health surveillance by enabling the integration of epidemiological, clinical, environmental, mobility, genomic, and digital data to detect emerging threats and anticipate disease transmission. However, the expansion from detecting outbreaks to predicting individual or group-level behaviors raises important ethical concerns related to privacy, autonomy, discrimination, transparency, accountability, and public trust. This mini-review examines the ethical limits of AI-powered public health surveillance, with particular attention to the transition from population-level epidemic prediction toward individual-level inference and profiling. It highlights how extensive data integration, algorithmic bias, unequal predictive errors, opaque models, and mission creep can undermine both equity and the legitimacy of surveillance systems. The review argues that predictive capability alone cannot establish ethical legitimacy. Instead, AI surveillance should be governed by necessity, proportionality, purpose limitation, data minimization, equity, transparency, meaningful human oversight, and mechanisms for accountability and redress. A lifecycle governance approach is further required to ensure continuous assessment of data quality, subgroup performance, model drift, unintended harms, and changing public-health needs. Ultimately, the ethical boundary of AI surveillance should be defined not by what can technically be predicted, but by what can be predicted and acted upon without compromising fundamental rights, equity, and public trust.