AI Is Entering Clinical Practice Faster Than Medicine Can Adapt

Artificial intelligence is becoming part of everyday clinical practice, but evidence on how physicians should use it safely is still catching up.

That tension is highlighted in a recent article by The Atlantic, which brings together several new studies on AI-assisted clinical reasoning and the risks of premature adoption.

Strong performance does not guarantee safe use

In a study published in Science, researchers from Harvard and Stanford found that ChatGPT outperformed hundreds of physicians on complex diagnostic cases.

Despite the results, lead author Adam Rodman, MD, emphasized that the study was an evaluation of diagnostic reasoning, not evidence that clinicians should rely on AI in routine practice.

A separate randomized trial published in NEJM AI reached a different conclusion. When physicians worked with intentionally incorrect AI recommendations, many incorporated the errors into their own clinical reasoning instead of identifying them.

AI is already part of the workflow

More than 80% of physicians report using generative AI professionally in some capacity.

At the same time, many clinical reasoning tools remain outside formal regulatory review, and hospitals continue introducing new AI applications into daily practice.

Organizations are also raising concerns. ECRI ranked inappropriate chatbot use among the leading health technology hazards for 2026, reflecting growing attention to the risks of overreliance and misuse.

The next challenge is training

Medicine has decades of experience teaching clinicians how to interpret imaging, laboratory results, and clinical guidelines. AI requires a different skill: knowing when not to trust the answer.

For surgeons and physicians, the question is no longer whether AI will become part of clinical practice. It already has.

The challenge is ensuring that clinical judgment evolves as quickly as the technology.

Why It Matters

The biggest barrier to safe clinical AI may no longer be the software. It may be preparing clinicians to use it appropriately.

Source: The Atlantic