The Risk Hiding in Generic AI

It is 7 p.m., the schedule ran late, and a complex patient left a question. The doctor opens a generic AI assistant and types the case — name, age, history, "what is the best course of action?" It feels natural. The tool is fast and helpful.

The problem is not using AI. It is which AI — and what happens to the patient's data the moment it leaves your screen. With a consumer-grade assistant, identifiable health information leaves the consulting room and lands on a technology company's servers, under consumer terms of use, outside your control. That is where the risk begins — in São Paulo, Lisbon, London or New York alike.

Generic AI was not built for the clinic

Tools like ChatGPT, Gemini or the consumer version of Claude are remarkable — but built for general use, under consumer terms. They do not contractually take on the duties the law and medical ethics place on anyone handling health data. Three differences matter:

• No health-data contract. A real clinical system runs under a formal agreement with its technology provider — a Business Associate Agreement under HIPAA in the US, an operator contract under GDPR/LGPD in Europe and Brazil — binding the provider to protect the data, never use it for its own purposes, and report breaches. Consumer AIs generally do not offer this to the ordinary user. Without it, using patient data is itself a violation.
• The data can train the model. In many consumer products, what you type may be retained and used to improve the system. Your patient's record could, in principle, become training material — incompatible with confidentiality.
• No record, no audit, no access control. A chat is not a medical record. It keeps no audit trail, guarantees no retention, and does not isolate one patient from another.

What the doctor is really risking

Each risk maps directly to the law the doctor already answers to:

• Breach of medical confidentiality — a disciplinary matter before the medical board, anywhere in the world.
• Unlawful processing of sensitive health data — Protected Health Information under HIPAA, a special category under GDPR (art. 9), sensitive data under LGPD (art. 11) — often with cross-border transfer to jurisdictions without adequate safeguards.
• No record of the AI's use. The rules for AI in medicine converge on two points: the decision is the doctor's, and the use must be documented (CFM Resolution 2.454/2026; the human-oversight and traceability duties of the EU AI Act; GDPR art. 22, against solely-automated decisions). A loose chat leaves no record.
• Liability for a wrong suggestion. Generative AI can be confidently wrong. Responsibility for the clinical conduct is never the machine's — it is the doctor's.

How KDoctor protects the doctor

KDoctor was built from the ground up as a clinical platform — not a generic chat with a medical veneer. The AI stays fast and useful, but the patient's data never leaves clinical control, and every duty above has a matching safeguard:

• The data never goes to an off-the-shelf AI. KDoctor's AI runs under a formal health-data contract (BAA / operator), on covered, auditable infrastructure, contractually forbidden from training on the data. The patient never becomes training material.
• Minimization and identity protection. Only what is clinically necessary is sent; the patient's identity is protected before processing; and where it can run locally, the data never crosses the open internet.
• The doctor always decides. AI is a copilot, not a pilot — every suggestion is presented for the doctor to accept, modify or reject, and that decision is recorded.
• Complete traceability. Every AI interaction is logged in the patient's record — what was used, what was suggested, what the doctor decided — satisfying the CFM, HIPAA's audit trail and the EU AI Act's logging duty at once.
• A real medical record with retention, integrity, and a trail of who read or changed each piece of information.
• Each patient isolated, every access logged — enforced in the database, not just on the screen.
• Curated clinical knowledge with sources — specialized, validated, and always under the doctor's review: the opposite of an untraceable generic answer.

Generic AI vs KDoctor

The bottom line: with generic AI, the doctor lends the patient to a technology company. With KDoctor, the AI comes to work inside the consulting room — under confidentiality, under record, and under the doctor's command, to the standard of the US, Europe and Brazil at once.