An AI should not mention cancer merely because it appears somewhere in a possible differential, or conceal it simply because the word could upset someone. The communication should reflect the clinical purpose, the available information, the patient's concerns and what happens next. AMIE's study offers a specific reason to examine that balance, not a universal rule about when to name cancer.
In its release dated 9 October 2026, Beth Israel Deaconess Medical Center describes one clinician rating an interaction as somewhat harmful because mentioning lymphoma might have caused anxiety. That was a reported clinical concern. It was not proof that AMIE caused a measured increase in patient anxiety.
What the reported episode can and cannot tell us
The episode matters because it concerns the effect of diagnostic language, not simply whether a possible diagnosis could be medically defended. A term can be relevant to a clinician's reasoning while still requiring careful explanation to a patient.
However, the institutional report does not provide the basis for reconstructing a complete consultation, assigning a probability of lymphoma or determining what the patient ultimately felt. It would be inappropriate to turn the account into a claim that AMIE frightened a patient into unnecessary treatment, or that the system should never discuss serious diagnoses. The October 2026 report supports neither conclusion.
The more useful question is what a patient understood after hearing a possibility and whether the subsequent clinical conversation resolved uncertainty. That question should guide the evaluation rather than the emotional force of the word cancer.
A differential diagnosis and a patient explanation serve different purposes
A clinician's working differential can include possibilities that are unlikely but consequential, competing explanations that need separating and provisional labels that may change after examination. It is a tool for organising decisions.
A patient explanation has additional tasks. It must convey what is known, what remains uncertain, why a possibility is relevant and what the patient should do with the information. Delivering the entire internal list without context does not necessarily achieve those tasks.
This is consistent with the GMC's decision-making and consent guidance, checked on 10 October 2026. It emphasises dialogue, information relevant to the individual and honesty about uncertainty. It does not support withholding important information merely because it may be upsetting, or replacing a conversation with an indiscriminate list.
The distinction is especially important before an appointment. A possibility introduced during intake may shape the concern that the patient brings to the clinician. The receiving professional needs to know not just the generated differential, but what the patient heard and what they now believe it means.
A fictional communication exercise
Consider a constructed teaching scenario involving a patient asking about a persistent lump. No details are taken from an AMIE participant, and the wording below is not an AMIE transcript.
An uncontextualised explanation might say: "Possible causes include infection, inflammation and lymphoma." That sentence names categories but leaves the patient to infer their relative likelihood, the limits of the assessment and the urgency of any action.
A more useful teaching alternative might be: "There are several possible explanations, and this conversation alone cannot establish the cause. Your clinician needs to assess the lump and the rest of your history before deciding whether any investigation is needed. A serious cause may need consideration, but listing one is not the same as diagnosing it. What are you most concerned this could be?"
This is not a script suitable for every presentation. In an actual encounter, the information available could justify a more direct explanation, a different level of urgency or a different route to care. The exercise asks learners to improve the structure of uncertainty, not to soften every serious discussion into reassurance.
The next teaching step is to ask what information would change that wording. Would a new examination finding justify more explicit concern? Has the patient already named cancer? Is there a previous investigation or specialist assessment that should shape the explanation? A sound answer depends on those facts, not on choosing a universally reassuring tone.
Avoid false precision as well as vague reassurance
A list can make remote possibilities appear equally likely. An invented percentage can be worse, giving a precise-looking probability without a validated basis for this patient. The corrective move is not to manufacture a number. It is to explain why the current information is insufficient and what assessment could reduce the uncertainty.
Likewise, "it is probably nothing" can be unhelpful when the speaker has not established enough information to justify reassurance. Clear uncertainty can coexist with a clear action: the diagnosis remains unresolved, but the next step and responsibility for it can still be specified.
The GMC's dialogue guidance provides the professional framework for tailoring information and exploring what matters to the patient. It does not validate any particular AI-generated wording or convert this educational example into clinical advice.
What should a patient-facing communication study measure?
A useful evaluation would ask patients to explain, in their own words, what the AI said was known, what remained uncertain and what should happen next. It would assess whether serious possibilities were mistaken for confirmed diagnoses and whether reassurance was understood as permission to ignore a changing situation.
It should also distinguish distress from inappropriate communication. A difficult but necessary conversation may cause understandable concern. A pleasant interaction could still leave the patient misinformed. The objective is not to make every conversation emotionally effortless; it is to support informed, proportionate understanding.
For a pre-visit system, the evaluation could follow the concern into the appointment. Did the clinician receive the relevant context? Was the patient's interpretation clarified? Did the encounter generate avoidable repeat contact or help the patient ask a question they otherwise would not have raised? These are proposed endpoints, not findings from the AMIE study.
Practising the transition from reasoning to explanation
A useful teaching session gives the learner a fictional differential, asks them to identify which possibilities need discussing and then asks them to explain their choice. Another learner or supervisor plays the patient and describes what they understood.
The success criterion is not whether the learner used a reassuring phrase. It is whether the explanation preserved uncertainty, addressed the patient's concern and connected the information to an appropriate next step. iatroX's reasoning and consultation-practice formats can support such educational work, but no claim follows that iatroX has solved patient-facing AI communication or reproduced this exact case in its catalogue.
Frequently asked questions
Did AMIE cause anxiety by mentioning lymphoma?
The 9 October 2026 BIDMC report describes a clinician's concern that this might have happened. It does not establish a measured anxiety outcome caused by AMIE.
Should medical AI avoid mentioning cancer?
Not categorically. Important possibilities should be discussed honestly and in context, rather than hidden or presented as an unexplained list.
How can clinicians practise explaining an uncertain differential?
Use a fictional case, justify which possibilities need discussing and ask another person to explain back what they understood. Review the accuracy of the message and its connection to the next step, not just its tone.
Practise explaining the reasoning behind a clinical answer →