We can, but should we? Red flags for hasty, low integrity AI integration into healthcare

News reports in recent weeks highlight several potential missteps in the implementation of artificial intelligence in healthcare.A patient sued Open AI claiming that ChatGPT gave him harmful medical advice, a whistleblower claimed a major academic medical centre cut corners in evaluating patient facing AI tools, and a large insurer was accused of using reckless algorithms to prematurely discharge patients from postacute care.1Reports like these are likely to continue as healthcare organisations race to implement new AI tools.At the same time, public scepticism about AI is growing, which further raises the stakes of getting AI implementation wrong, particularly where patients are directly affected.2 Trustworthy AI implementation in healthcare entails not only ensuring outputs are accurate and reliable, but also that the processes by which healthcare organisations adopt, implement, evaluate, govern, and communicate about their AI efforts maintain their integrity, including proactively detecting and responding to problems.3As health systems pursue AI integration,…