Have a thorny medical question? Your doctor may be using AI for that

A groundbreaking new study published in Nature Medicine on June 12, 2026, reveals that general-purpose large language models (LLMs) like Google's Gemini and OpenAI's ChatGPT are outperforming specialized clinical AI tools, including the rapidly adopted OpenEvidence, across key medical benchmarks. This finding challenges the industry's belief in purpose-built AI, as OpenEvidence, a startup now valued at $12 billion, has seen its physician-led adoption surge to over half of U.S. doctors, who used it for 30 million consultations last month alone. The platform offers instant, evidence-based answers to 'thorny medical questions,' with Dr. Nicholas Gavin, Mount Sinai's Chief Clinical Innovation Officer, confirming its widespread, often 'shadow AI' use among his colleagues. This explosive, bottom-up adoption of tools like OpenEvidence has ignited a scramble among health systems and regulators to understand and control the influx of AI into life-or-death clinical decisions. Mount Sinai, for instance, is moving to integrate OpenEvidence into its Electronic Health Record (EHR) system, carefully navigating patient privacy concerns by initially restricting patient data access. However, the Nature Medicine study results complicate this integration, highlighting concerns about the clarity of specialized AI outputs, even if not more prone to 'hallucinations' than general LLMs. The U.S. Food and Drug Administration (FDA) continues to grapple with an outdated regulatory framework, struggling to oversee adaptive AI technologies and address critical issues like AI Bias in Medicine and the 'black box' problem, where AI decision-making lacks transparency. The immediate future demands a delicate balance: fostering innovation while implementing robust regulatory frameworks and rigorous, real-world validation of AI tools. Industry leaders and policymakers must now reassess procurement strategies and accountability mechanisms, particularly in light of the unexpected performance gap between general and specialized AI models. As competition intensifies—with legacy players like UpToDate also incorporating AI and new certifications emerging like CHAI's RUAIH for responsible AI—the evolution of AI in healthcare will hinge on establishing trust, transparency, and demonstrable patient benefit, pushing for a future where 'medical super-intelligence' truly frees clinicians for higher-value care rather than introducing unforeseen risks.