AI Has Genuine Power to Improve Public Health. Realizing That Power Requires Getting the Risks Right.

Let me start with my bias, because I think transparency is a precondition for analysis. I believe that artificial intelligence is a transformative tool that can help public health agencies avert illness and death in their communities, provided it is deployed responsibly and with genuine attention to the ethical complexities it introduces. That belief is grounded in what I have seen AI systems begin to accomplish across a range of public health functions, from processing and interpreting large volumes of surveillance data that would overwhelm any human analyst working alone, to tailoring emergency communications to specific populations in ways that improve both reach and comprehension, to strengthening programs for immunizations, sexually transmitted infections, tuberculosis, and maternal and child health.

Public health has always been, at its core, a field defined by the problem of scale. The diseases and social conditions we work to prevent operate across entire populations, and the data we depend on for decision-making arrives in volumes, at speeds, and from sources that strain the capacity of traditional analytical methods. AI offers something genuinely valuable in that context: the ability to identify patterns in data that would otherwise remain invisible, to process information faster than any team of epidemiologists could manage manually, and to translate complex findings into communications tailored to the specific communities that need them most. These are capabilities with real consequences for real people, and dismissing them because AI also carries risks would be as intellectually irresponsible as embracing them uncritically.

The Tensions at the Frontier of AI Development

Earlier this month, a senior safeguards researcher at Anthropic — the AI company that has positioned itself as a safety-oriented actor in the race to develop increasingly powerful generative AI systems — resigned, warning in his resignation letter that the “world is in peril,” citing concerns about AI, bioweapons, and interconnected global crises. The New Yorker simultaneously published an in-depth investigation into Anthropic and the tensions its employees navigate while trying to build advanced AI systems and protect against the risks those systems might pose to humanity.

What makes this moment unusual is that the concerns being raised are coming from inside the institutions building the technology, by people with direct knowledge of what these systems can and cannot do. That is a different kind of warning than the abstract philosophical objections to AI that circulated a decade ago. The researcher who resigned had led work specifically on reducing risks from AI-assisted bioterrorism and on understanding how AI assistants could distort human judgment in ways users might not recognize. His departure, and the concerns that prompted it, deserve more sustained public attention than the news cycle has given them.

Adding a layer of complexity that warrants its own analysis, the U.S. Defense Department recently took the position that Anthropic may represent a risk to the U.S. government — because the company is considered too focused on ethics and protecting human health. The implication embedded in that position, that a safety-oriented approach to AI development is itself a national security concern, reflects a set of institutional priorities that public health professionals and biosecurity experts should examine carefully.

A Tool That Can Help and a Tool That Can Harm, Simultaneously

The honest answer to questions about whether AI will benefit or endanger public health is that it will do both, and the ratio between those outcomes depends heavily on choices being made right now by companies, governments, and research institutions. The same capabilities that make AI valuable for vaccine design and outbreak detection — its ability to synthesize scientific literature, solve complex technical problems, and operate across biological domains at speed — are the capabilities that raise legitimate biosecurity concerns. Those dual-use tensions cannot be resolved by choosing sides in a debate about whether AI is good or bad. They require the kind of layered, evidence-based risk management that public health has applied to other dual-use technologies throughout its history.

What I find most clarifying about the current moment is that the risks being identified by credible researchers are neither science fiction nor distant hypotheticals. They are grounded in demonstrated capabilities, operating in a regulatory environment that has not yet developed adequate frameworks for managing them. Public health agencies have a long track record of preparing for rare catastrophic events while simultaneously addressing everyday harms — that dual focus is, in many ways, the defining operational challenge of the field. Applying that same framework to AI, with the urgency the technology’s pace of development demands, is among the most important tasks facing public health institutions today.