Oct 23, 2025
3 min read
The European Society for Medical Oncology is publishing the first guidelines on the responsible use of artificial intelligence in oncology. Priority: protecting patients and doctors
Chatbot, symptoms and diagnosis: esmo breaks the rules for AI in oncology
The European Oncology Society publishes guidelines for the effective use of languages.Three areas of work, one thing: Protecting patients and doctors
Chatbots that talk to patients, algorithms that scan medical records, virtual assistants that help doctors make decisions.This is not science fiction, but the arrival of artificial intelligence in the classroom.But while the contribution from one side is noticeable, there's one question that worries experts: Who will be in charge of the AI when it does it all?This question was answered by ESMO, the European Society of Medical Oncology, at its annual congress in Berlin.
Ai in oncology: the basic rules of its safe use
At a time when generative artificial intelligence communicates, cares and decides, ESMO has made a clear choice: it has set the rules.These have been developed for oncology so far and aim to protect patients and doctors from invisible but concrete risks.The document is called ESMO guidance on the use of major language models in clinical practice and was published only in the journal Annals of Oncology.Twenty-three recommendations, three applications, a basic principle: artificial growth intelligence, should never replace, clinical judgment.
AI in Cancer Page 3: Patients, Physicians, Institutions
"Innovation must bring real benefits to patients and concrete solutions for Clinicia - says Fabrice André, we offer a common framework for the integration of AI without sacrificing responsibility, transparency and data protection."The document is divided into three contexts of use, each with different needs and risks.
AI for Patients: Not Just Chatbot Answers
In the first level of application, the one intended for patients, artificial intelligence takes the form of chatbots and virtual assistants that are capable of communicating, explaining the diagnosis, helping to solve symptoms or orienting towards treatment paths.But behind the promise of accessibility and speed lies a huge risk, the risk of errors.The ESMO guidelines reiterate this firmly: any patient application it must include is monitored on the clinical pathway, where the physician remains the point of reference.AI can facilitate, explain, monitor, but not decide.And above all, it must respect a non-negotiable rule: strict protection of personal data.In other words, AI is not a shortcut.It is a tool that can improve the patient experience only if it is inserted into a responsible clinical ecosystem, where every reaction is monitored, every data is protected, and every decision remains in the hands of those with expertise.
Chatbot for doctors, nurses and healthcare professionals
Of course, doctors and nurses can also use language models to translate reports, write documents or search for information.Also in this case it is essential that each application is formally verified, its boundaries are declared in a transparent way and the final responsibility is always left to the individuals.“AI can save time and errors,” says Jakob N. Kather, oncologist and vice-president of the ESMO task force on digital health.“It can help reduce it, but it can't decide for us,” he said.
AI for hospital systems
Here, the artificial intelligence it inherits from the systematic hospital, based on the medical record, the data or the patient clinical trial based on the database.But it requires continuous monitoring and identification, prior to inspection, even if the process or data source changes.
Its value depends on who uses it
A crucial point is about the quality of initial data.Furthermore, artificial intelligence is not infallible, but it is a mirror of what it receives.If, in fact, the information is incomplete or the questions are formulated, even the most advanced AI can produce incorrect solutions.And in oncology this is not a detail.For this reason, the guidelines require consistent supervision and transparency.All applications you must clearly identify what you can and cannot do.Each answer must be verified.
Help, not orders: support, not ordered
ELCAP focuses on a specific type of artificial intelligence: auxiliary intelligence.These are models that help professionals find information, write texts or analyze data, but always under human control."These systems improve, not replace, workflows and clinical decisions," Kater repeated.But Esmo does not ignore the future.Models that can operate independently, without direct input, are emerging quickly.Tips.And they raise new questions: Who runs them?How are they regulated?How is security guaranteed?"We will need specific recommendations," admitted Andre.For now, let's build the first dam. ”
Common denominators: Trust is built together
In oncologia, come in tutto l’ambito salute, ogni decisione è delicata. E ogni tecnologia deve essere all’altezza. «L’uso responsabile dell’intelligenza artificiale richiede standard condivisi – ha concluso André. – Elcap definisce come implementare modelli linguistici in modo da migliorare la qualità, l’equità e l’efficienza dell’assistenza oncologica, senza compromettere la fiducia nel giudizio clinico». ESMO, insomma, in un’epoca in cui l’AI promette di rivoluzionare la medicina, sceglie la via della cautela intelligente: non frenare l’innovazione, ma incanalarla in percorsi sicuri, trasparenti e umani. Perché in oncologia, più che altrove, ogni parola conta e ogni decisione può fare la differenza.
