- A series of medical AI tools launched in 2026, such as Copilot Health (Microsoft), Health AI (Amazon), ChatGPT Health (OpenAI), and Claude, allow users to ask health questions and access personal medical records.
- Huge demand: Copilot alone receives up to 50 million health queries daily, showing that AI is becoming the “first doctor” for many people.
- AI has the potential to reduce the burden on the healthcare system by supporting triage, helping patients decide whether they need an in-person visit.
- However, a study at Mount Sinai found that AI sometimes recommends over-treatment for mild conditions and misses emergencies, raising safety concerns.
- Experts warn that widespread deployment without independent testing is a major risk, as corporate self-assessments may contain “blind spots.”
- Users without medical knowledge may provide incorrect information or misinterpret AI responses, leading to inaccurate diagnoses.
- Benchmarks like HealthBench show that AI has significantly improved but is still not perfect, especially in asking follow-up questions to users.
- Some experimental systems like Google’s AMIE achieved accuracy comparable to doctors in studies but have not been released due to safety and fairness concerns.
- While AI doesn’t need to be perfect, the big question remains: whether the benefits outweigh the risks in practice.
📌 Medical AI is growing fast with tens of millions of queries per day, opening opportunities to improve healthcare access. However, current evidence is insufficient to ensure safety, particularly in diagnosis and emergency handling. Without large-scale independent testing, AI can be both a solution and a risk. The future depends on balancing rapid deployment with ensuring reliability in sensitive medical environments.
