OpenAI has announced that ChatGPT Health is now available to all users over 18, across all plans including the free tier. While the news sounds promising, from a technical and business perspective it is worth analyzing the real implications. This is not just a new conversational feature: it involves handling sensitive health data, which demands far higher standards of privacy and reliability than a generic chatbot. My skepticism is not unfounded: I have seen how the speed of release in the artificial intelligence sector often sacrifices the maturity required for critical environments.
The most worrying change is the backward step in privacy protection. Originally, OpenAI promised that health conversations would not be used for training and would be isolated from others. Now the terms have shifted: only conversations where users explicitly connect their medical records or Apple Health data are considered protected. This means that if you simply describe a symptom without attaching documentation, that conversation does not enjoy the same guarantees. In practice, to obtain real privacy you have to share more data, not less. A dangerous paradox that any serious software development company would avoid. At Q2BSTUDIO we know that cybersecurity and transparency are non-negotiable pillars when handling sensitive information.
OpenAI boasts about its HealthBench evaluator to claim that ChatGPT outperforms real doctors. However, a technical analysis reveals serious limitations. HealthBench uses rubrics designed by physicians to rate short, isolated responses, but it does not simulate real conversations where patients change topics, use colloquial expressions, or present rare conditions. It also does not assess continuity of care or management of multiple specialists. Furthermore, the rubrics themselves show suspicious duplications: the same response receives points for 'advising to seek emergency care' and then more points for 'stating to contact emergency services.' This looks more like a trick to inflate scores than a rigorous evaluation. In the world of custom software development, any metrics system must be validated with real data and varied usage scenarios.
Beyond controlled tests, the cases of real harm are undeniable. A recent lawsuit from a pastor who nearly died after following advice from ChatGPT-4o about a pulmonary embolism; other lawsuits link the chatbot to suicides and psychosis. The medical safety organization ECRI has already named the misuse of AI chatbots as the top health technology hazard for 2026. This is not an exaggeration: when a tool generates responses with an appearance of authority, users tend to trust blindly, especially if the system claims to have 'connected your medical data.' The line between an informational assistant and a medical substitute is blurred, and OpenAI is not doing enough to mark it.
From a business perspective, OpenAI's approach recalls the rush to launch minimally viable products without proper verification. In contrast, companies like Q2BSTUDIO bet on development based on real needs, with exhaustive testing cycles and regulatory compliance (GDPR, HIPAA where applicable). Instead of relying on a generic language model to interpret diagnoses, the sensible approach is to combine artificial intelligence with secure cloud architectures (like AWS or Azure) and Business Intelligence systems (Power BI) that allow visualization and auditing of information. Building specialized AI agents for healthcare requires a design focused on the clinical domain, not simple prompts tuned with questionable rubrics.
OpenAI offers some tips to mitigate risks: use temporary chats, review imported data, disconnect accounts with a 30-day deletion option, and require permission every time records are accessed. But these are all patches on a fundamental problem. The company admits it is 'still testing' additional safeguards, which indicates there are no solid guarantees. If you decide to use ChatGPT Health, do so with extreme caution, understand it is not a medical diagnosis and never replaces the judgment of a professional. Remember, as a cybersecurity expert once said: 'privacy is not a feature, it is a responsibility.' And in the healthcare field, that responsibility cannot be delegated to a chatbot without human oversight.
In conclusion, the arrival of ChatGPT Health for all users represents an advance in accessibility, but also a calculated risk that OpenAI takes too lightly. Companies that truly understand the healthcare sector know that technology must serve the patient, not the other way around. That is why at Q2BSTUDIO we recommend exploring BI and Power BI solutions for clinical data management, cloud platforms with granular access controls, and AI agents trained on specific datasets validated by professionals. Only then can trust be built in a field where an error costs lives.
