“Physicians for AI Safety” invites you to sign up

Physicians are trained to prevent what they cannot cure. As AI becomes increasingly powerful, the founders of Physicians for AI Safety believe the medical community has an important role to play in preventing severe and irreversible harms. They’re building a coalition of physicians, residents, and medical students who want these risks taken seriously.

If this resonates with you, they’re inviting you to sign onto a simple statement:

“Preventing severe harm from AI should be a global priority, alongside pandemics and nuclear war”

Why this initiative?

Physicians have spoken out before when threats to health were underestimated, as they did on nuclear war and climate change. Advanced AI is now such a threat. Many physicians worry that powerful AI could cause severe harm, but often find it hard to say so publicly. This one-sentence statement makes that easier, and shows that a large number of physicians see advanced AI as an enormous threat to public health that must be urgently addressed.

Why physicians?

Physicians are among the most trusted professions around the world. Governments listen when doctors speak, and that gives physicians influence over policy decisions. And they are trained to think about population health risks as well as individual patients.

Physicians have done this before. In 1980, doctors from the United States and the Soviet Union founded International Physicians for the Prevention of Nuclear War (IPPNW). They argued that there would be no one left to treat after a nuclear war, so the only medicine is prevention. The movement received the Nobel Peace Prize in 1985. In 2021, over 200 health journals published a joint editorial calling for emergency action on climate change. Now, the same argument may apply to advanced AI. We must act now.

What are the health risks associated with AI?

Physicians for AI Safety focuses on harms from AI that is misused or out of human control, and that are severe, widespread and almost impossible to reverse. These include:

• Attacks on critical systems such as hospitals, power grids and water supplies

• Biological weapons made easier to design, produce or disseminate with AI assistance

• Military use that increases casualties as weapons become faster, automated and easier to turn on civilians

• Concentration of power through mass surveillance, loss of human agency and the weakening of democracy

• Loss of human control over increasingly autonomous systems that pursue goals different from ours

Richard Armitage, a GP in the United Kingdon and one of the three founders behind Physicians for AI Safety, expands on several of these risks to health in a 2025 commentary in the Journal of Evaluation in Clinical Practice. Automation is projected to cost tens to hundreds of millions of jobs within a decade, and unemployment brings well-documented harms to health, including depression, harmful drinking and suicide. As AI systems become more capable, cyberattacks using them could cripple health systems by altering electronic health records or interfering with patient monitoring and diagnostic equipment. The same attacks could hit power grids and other critical infrastructure. Advanced AI could also help unregulated actors obtain chemical, biological, radiological or nuclear weapons, and speed the development of more lethal ones such as new engineered pathogens. AI-driven disinformation could spread credible but false health information that undermines trust in health systems and reduces uptake of vaccination and screening. Federspiel and colleagues reached similar conclusions two years earlier in BMJ Global Health, writing with co-authors from IPPNW, and they add detail on military use. Lethal autonomous weapons could be mass-produced cheaply, and a single shipping container could hold a million explosive-carrying drones programmed to kill without human supervision.

Armitage also takes his examples of loss of control from health. A misaligned system pursuing a legitimate goal, such as better maternal health or infectious disease control, might do so with no regard for human rights, through forced sterilization or universal quarantine. It might also seek power, resist being shut down, or appear aligned during training and behave differently once deployed. Beyond a certain level of autonomy, its behavior would likely be impossible to reverse.

This is no longer a fringe view in global health. The new Lancet Commission on 21st century global threats to health, published in October 2026, lists malicious use of AI among its 17 major threats to health this century. Its account covers much of the same ground, from AI-designed pathogens and ransomware attacks on health facilities to loss of control. It is one of only two threats, alongside nuclear war with nuclear famine, that the Commission judges to carry a risk of human extinction.

AI also raises important questions inside medicine, like diagnostic error and algorithmic bias. However, Physicians for AI Safety focus on harms to entire populations that might never be undone.

What Physicians for AI Safety is not

They are not against all forms of AI. They use AI in their work and are excited by the benefits it can bring to patients. They oppose the deployment of powerful systems without sufficient safeguards. They are independent of any AI company or political party, and every signatory signs in a personal capacity.

Who are the founders?

Dr Erik Jentzen, A German physician based in Switzerland working on global catastrophic biological risks. He is Director of the Zurich Biosecurity Hub and a board member of High Impact Medicine.

Esha Khurana, a psychiatrist (MD, MPH), psychotherapist, and Assistant Director of Behavioral Health at UPMC. Her clinical and academic work has included psycho-oncology, women's behavioral health, and psychosomatic medicine, with a broader focus on the intersection of mental health and public health.

Dr Richard Armitage, A British General Practitioner, DPhil student in Clinical Medicine at the University of Oxford, biosecurity researcher, Honorary Clinical Assistant Professor at the University of Nottingham, and Associate Editor of the Journal of Evaluation in Clinical Practice.