As artificial intelligence advances, society is asking a necessary question: are we moving too fast? In some areas, that concern is justified. Powerful and generally applied frontier models require rigorous evaluation, thoughtful governance and clear accountability. Much of the AI used in healthcare is designed for defined purposes within clinical workflows, distinct from the more autonomous, general-purpose systems driving the broader debate.
Patients cannot afford reckless AI deployment, but neither can they afford blanket hesitation when health systems are under pressure, clinicians are stretched, and too many people wait too long for diagnosis or treatment. Instead, healthcare needs the right AI for the task, validated where it will be used and governed according to its risk.
Standing still carries risk, too
In healthcare, safety is not only about preventing new harm. Patients also face risks when care is delayed, expertise is unevenly distributed, and clinicians are stretched.
“In imaging, responsible AI can help clinicians manage growing demand by improving image acquisition and quality and reducing workflow friction.”
Breast cancer screening offers a timely example. As we enter Breast Cancer Awareness Month, a recent AACR report highlights the growing role of AI in early detection. In a randomized trial of more than 105,000 women in Sweden, AI-supported mammography performed at least as well as standard screening within a predefined margin; earlier results from the same trial showed a 44 percent reduction in radiologists’ screen-reading workload. A large UK study also found that AI-supported screening maintained or improved performance across subgroups defined by ethnicity, age and breast density.
The evidence is promising, but it also underscores the need to validate AI across populations and care settings. Responsible adoption requires weighing both sides of the equation: the risks of moving too quickly and the potential costs of waiting when evidence supports meaningful benefit.
At the same time, AI will not improve healthcare simply because it is deployed. If a system performs well only in highly resourced environments, depends on infrastructure that is unavailable elsewhere, or is trained on data that do not reflect the people being served, it can widen the very gaps it is intended to close.
“The question, then, is not whether healthcare should move forward with AI. It is how we move forward with the right level of evidence, safeguards, and oversight for the risk involved.”
Regulation should be risk-based and clinically grounded
Not every AI system presents the same level of risk, and policy should not treat it as though it does. An AI tool that helps a clinician acquire an ultrasound exam or improve image quality does not carry the same potential consequences as a system that generates a diagnosis, recommends treatment or takes action across a clinical workflow. The evidence, testing and safeguards should rise with the potential harm and degree of autonomy.
In healthcare, autonomy has to be earned by evidence. The FDA’s Predetermined Change Control Plan framework offers one model, allowing predefined changes within validated boundaries while maintaining appropriate oversight.
Evidence must also extend beyond performance on a test set. We need to understand whether AI delivers meaningful clinical or operational benefit in the environments where it will actually be used, including across different patient populations, care settings and levels of resources.
“That is how healthcare can move both safely and at speed: by enabling responsible innovation where risks are understood and manageable, while applying greater scrutiny where the clinical consequences are highest.”
The bottom line
Ultimately, responsibility is shared: regulators must set proportionate rules, developers must validate AI for the settings and populations where it will be used, and health systems must monitor how it performs in practice and be able to intervene, roll back, suspend, or withdraw a solution when needed.
We see our responsibility to work across healthcare to bring safe, dependable AI into care while continuing to innovate. Patients deserve AI they can trust, and they’re counting on us to drive care forward, moving from treatments toward cures.
