AI wants to become our first medical consultation. Are we ready?

Artificial intelligence is accelerating its entry into one of the most sensitive and regulated domains: healthcare. In the space of a few days, two of the biggest global players in the field - OpenAI and Anthropic - presented their dedicated tools, ChatGpt Health and Claude for Healthcare respectively, marking a new phase in which chatbots are no longer limited to providing generic information, but integrate personal health data and real clinical systems. In parallel, while large language modelling companies set themselves up as allies of patients and doctors, Google was forced to withdraw part of its Ia-generated health summaries after a journalistic investigation found misleading and potentially dangerous indications. This episode is a counterpoint to the new race: huge opportunities, but also real risks.

The announcement of ChatGpt Health ushered in the new season of Ia applied to wellness. The tool creates a space for medical questions and allows users to link medical records, lab results and fitness apps, promising support in making sense of information that is often scattered, technical and difficult to interpret. OpenAI explicitly warns that ChatGpt Health is not a diagnostic system and does not replace the doctor, positioning it as a 'health navigator' rather than a 'virtual doctor'.

A few days later came Anthropic's response. With Claude for Healthcare, the company introduces both functionality for patients and tools designed for hospitals, insurers and companies in the regulated medical sector. Claude for Healthcare can in fact: access reliable healthcare and insurance databases; simplify clinical documents; verify insurance coverage; support complex administrative paperwork. For consumer users, on the other hand, it offers functions similar to ChatGpt Health: linking to personal health records, Apple Health or Android Health Connect, identifying trends in the data and preparing questions for doctor visits. According to Anthropic, health data is not used to train models and can be disconnected at any time, a sign of how central the issue of health privacy has become to the legitimacy of these tools.

The third piece of news of the week, seemingly distant, is one that casts a shadow over this new enthusiasm: Google has removed its 'AI Overviews' from some health searches, after an investigation by the Guardian showed errors and misleading simplifications. Among the examples cited: liver test values presented as 'standard' numbers devoid of clinical context; dietary advice for cancer patients contrary to medical guidelines Specialists warned that, in medicine, even technically correct information can become damaging if isolated from context: the exact problem that the automated summaries showed.

In this scenario, the question is not only whether Ia can respond correctly, but how, when and with how much responsibility it is integrated into the healthcare pathway. For now, both OpenAI and Anthropic exclude Ia from the diagnostic process and keep a physician 'in the loop'. However, the fact that the models can: access clinical data, interpret reports, synthesise medical guidelines and support insurance interactions, indicates that we are rapidly approaching a new digital health infrastructure in which the patient's first point of contact may no longer be human. On the consumer front, trust will be decisive: many users perceive these tools as authoritative, even when they should not. On the clinical front, on the other hand, the challenge is regulatory and medico-legal: who is responsible if the Ia gets it wrong? The race is just beginning. The technology promises tangible benefits - time savings, accessibility, understanding - but the question running through the industry remains the same: in health matters, can we really afford to trust artificial intelligence?