Richard Armitage is a GP, DPhil in Clinical Medicine student at the University of Oxford, Honorary Clinical Assistant Professor, and Associate Editor of the Journal of Evaluation in Clinical Practice.
Esha Khurana is a psychiatrist working at the University of Pittsburgh Medical Center.
Erik Jentzen is a German physician working in Basel, Switzerland.
The capabilities of artificial intelligence (AI) systems are growing faster than the safeguards needed to prevent them from causing severe harm. Physicians need to act now, as the profession has done previously when threats to health were underestimated.
On 24 September 2026 the Australian Prime Minister disclosed that an OpenAI agent had gained unauthorised access to the Medicare Statistics Reporting Service portal on 18 June.1 The agent had been set the task of researching public medical spending. It encountered repeated blocks, found ways to circumvent them, retrieved both public and non-public information, and wrote files to the internal server.1 OpenAI found the activity in August during an internal review of misaligned model behaviour and told Services Australia on 10 September, by email to a public inbox.2,3 Three other systems, including the Australian Institute of Health and Welfare and the Victorian Department of Health, may have been affected. No personal information is believed to have been accessed.1
This is the first publicly reported case of an AI agent breaching a government system without being instructed to, though not the first incident of its kind. For example, in July, an internal OpenAI model under cybersecurity testing gained administrator-level access to the servers of the AI company Hugging Face, again without human instruction.4 In Australia, the developer did not restrain its agent and did not notice for around two months. The information the agent sought was health data. The incident shows that health infrastructure is now within reach of AI systems their developers cannot reliably control. The same behaviour directed at electronic health records, prescribing systems, clinical decision support systems, laboratory information systems, or patient monitoring devices would put patients at direct risk.
Its statement is a single sentence: “Preventing severe harm from AI should be a global priority, alongside pandemics and nuclear war.“
Near-term AI harms already amount to a public health problem.5 AI can produce errors in clinical recommendations, health misinformation, algorithmic bias, and breaches of confidentiality. Millions already turn to AI for companionship, emotional support, and advice, risking dependency, isolation, reinforced harmful beliefs, self-harm, and damage to child and adolescent development. Systems that persuade and predict at scale threaten human agency through manipulation, surveillance, loss of privacy, and the concentration of power. Rapid disruption to employment, information systems, critical infrastructure, and social trust would harm population health, as would widening inequality. Autonomous weapons would remove human judgement from decisions to kill on the battlefield, and in time the same systems could be turned on civilians. AI systems that could deepen knowledge and innovation in the life sciences could also assist bad actors in the design, production, or dissemination of biological weapons, increase the chance of engineered pandemics, and produce pathogens that are resistant to all existing medical countermeasures. As demonstrated by the Australian incident, AI systems can now pursue their goals by harmful means, evade safeguards, and behave in ways their developers did not intend or predict. More advanced AI could escape human control entirely, with severe, irreversible, or even existential consequences.6,7
There is strong precedent for physician advocacy on societal-scale risks to health.
There is strong precedent for physician advocacy on societal-scale risks to health. In 1980, doctors from the USA and the Soviet Union founded International Physicians for the Prevention of Nuclear War, arguing that a nuclear war would leave no patient to treat, so the only suitable medicine was prevention. The movement received the Nobel Peace Prize in 1985.8 In 2021, over 200 health journals published a joint editorial calling for emergency action on climate change.9 Advanced AI now poses a comparable threat to health, and our profession has the same duty to speak.
We have therefore launched Physicians for AI Safety,10 an independent initiative founded in September 2026 by three physicians from the USA, Germany, and the UK. Its statement is a single sentence: “Preventing severe harm from AI should be a global priority, alongside pandemics and nuclear war.” This is not a statement against all forms of AI. We oppose the deployment of advanced AI systems that lack proven safeguards against harm. We support rigorous AI safety research, evaluation of AI capabilities and potential harm by independent third parties, safeguards against harmful misuse, transparency about capabilities and safety incidents, and urgent, effective national and international governance. The statement is deliberately short to allow physicians to voice their concerns without endorsing any particular AI forecast or policy. Its purpose is to show that many physicians see advanced AI as an enormous threat to public health that must be urgently addressed. It is open to physicians, residents, and medical students in any country. We ask readers to sign it at www.physiciansforaisafety.org and to share it with colleagues.
Physicians have a duty to speak when their patients are at risk. Advanced AI could cause severe harm to people and to entire societies. We must act now.
References
1. The Prime Minister of Australia, The Hon Anthony Albanese MP. Press conference – New York [Internet]. 2026. Available from: https://www.pm.gov.au/media/press-conference-new-york [accessed 9/10/26]
2. Handley E. OpenAI hacked Medicare portal, Prime Minister Anthony Albanese says [Internet]. Available from: https://www.abc.net.au/news/2026-09-24/ai-agent-accessed-australian-government-site-pm-says/107189078 [accessed 9/10/26]
3. Moore H, McMahon L, Vallance C. Why did an OpenAI system hack Australia’s health system – and can it be stopped in the future? BBC News [Internet]. 2026 Sep 24. Available from: https://www.bbc.co.uk/news/articles/cw24jm9rryy3o [accessed 9/10/26]
4. OpenAI. The Hugging Face incident and the road ahead [Internet]. 2026. Available from: https://openai.com/index/hugging-face-incident-and-the-road-ahead/ [accessed 9/10/26]
5. Armitage RC. The case for near-term artificial intelligence risks to be considered a public health problem. Global Public Health. 2026 Dec 31;21(1):2716984. doi:10.1080/17441692.2026.2716984
6. Armitage RC. Artificial General Intelligence and Its Threat to Public Health. Evaluation Clinical Practice. 2025 Sep;31(6):e70269. doi:10.1111/jep.70269
7. Federspiel F, Mitchell R, Asokan A, Umana C, McCoy D. Threats by artificial intelligence to human health and human existence. BMJ Glob Health. 2023;8(5):e010435. doi:10.1136/bmjgh-2022-010435
8. International Physicians for the Prevention of Nuclear War. Our Mission [Internet]. Available from: https://www.ippnw.org/about#mission [accessed 9/10/26]
9. Atwoli L, Baqui AH, Benfield T, Bosurgi R, Godlee F, Hancocks S, et al. Call for emergency action to limit global temperature increases, restore biodiversity, and protect health. The Lancet. 2021 Sep;398(10304):939–41. doi:10.1016/S0140-6736(21)01915-2
10. Physicians for AI Safety [Internet]. Available from: https://physiciansforaisafety.org/ [accessed 9/10/26]
Featured photo by Markus Spiske on Unsplash
