Many developers of medical artificial intelligence tools are unaware of key regulations governing their work, even though most believe they should be accountable for the systems they build, a multinational study has found.
Researchers at Singapore's Nanyang Technological University (NTU) surveyed 122 developers working on medical AI in Singapore, China, Hong Kong and the UK. The results, published in the journal npj Digital Medicine, show that only 57% of those questioned knew of at least one regulatory framework.
Awareness varied by seniority and workplace. Senior developers and those employed outside universities reported knowing more frameworks than junior colleagues and academic researchers. Two-thirds of respondents worked for organizations that had adopted no AI governance framework at all.
Responsibility is high, awareness is low
Despite the knowledge gaps, most respondents said they should be answerable for the systems they create. They viewed that responsibility as shared, with guidance expected from government or institutional ethics bodies, cooperation with users, and input from the organizations that supplied the data used in development.
Wilson Goh, an assistant professor at NTU's Lee Kong Chian School of Medicine and chief data scientist at the university's Centre of AI in Medicine, co-led the study. He said developers are well placed to judge data quality, the limits of their models, and risks including bias and hallucinations.
"The sense of responsibility among developers is reassuring," Goh said. He called the awareness gaps a concern, adding that "a more active approach is needed so AI models do not put patients or public confidence at risk."
Among the frameworks cited in the study are the European Union's AI Act and Singapore's Artificial Intelligence in Healthcare Guidelines. The researchers said limited familiarity with such safeguards could have consequences for patients and public trust once the tools reach clinical practice-a point especially relevant in countries investing heavily in digital technology.
Closing the regulatory gap
To address the issue, the researchers proposed that universities and training institutions teach regulatory frameworks as part of developer education. Inside companies, they suggested structured mentoring between experienced and junior staff to help newcomers understand compliance requirements.
They also called on national regulators to take a more active role in promoting compliance and to work toward aligning requirements across jurisdictions, drawing on the experience of developers themselves. For healthcare professionals looking to deepen their understanding of these governance issues, a learning path like the AI Learning Path for Regulatory Affairs Specialists offers structured training on regulatory compliance and policy automation relevant to medical AI.
Joseph Sung, NTU's senior vice-president for health and life sciences, director of the Centre of AI in Medicine, and dean of the medical school, who led the study, said the challenges of introducing AI into healthcare cannot be resolved by any single group alone. Collective responsibility for the consequences of deployment is needed to address patient safety, data privacy, and the long-term reliability of the technology in medicine, he said.
AI is being applied in more clinical tasks, from analyzing medical images to estimating a patient's disease risk. As healthcare systems increasingly adopt these tools, the researchers argue that improving developers' understanding of regulatory requirements will be as important as advancing the technology itself. For professionals already working in clinical settings, dedicated programs such as AI for Healthcare training can help bridge that gap.
Why this matters for healthcare professionals
If developers of medical AI lack full awareness of regulatory frameworks, the models they produce may carry undetected risks that become clinicians' problems. When a screening algorithm or diagnostic tool reaches a hospital, doctors and nurses rely on it being validated and compliant. The study suggests that improving regulatory literacy among AI developers cannot wait-delays in awareness mean potential safety issues will remain hidden until the tools are already in use with patients.
Your membership also unlocks:
