AI medication advice: A plausible answer is not a validated prescription
Language models can produce convincing errors. Define boundaries, approved sources and professional verification for medication-related questions.
Revisión editorial:
El texto de este artículo está disponible en inglés en esta versión del sitio.
Contexto y fuentes
OWASP identifies plausible misinformation and excessive reliance on LLM output as risks. Medication information needs a suitable current source and case-specific applicability; confident wording is not evidence. [1]
Software intended to support diagnostic or therapeutic decisions also requires medical-device assessment. MDCG guidance explains that MDR Rule 11 classification depends on the information's role and possible consequences, not a single class covering every chatbot or dosing tool. [2]
A disclaimer does not turn a general text assistant into validated dosing software. Conversely, saying language models can never calculate is too broad. Evidence of suitability for the actual intended use is what matters. [1], [2]
Alcance y límites
This article deliberately supplies no doses, conversions or individual medication advice. It does not predetermine liability or criminal responsibility.
Explicación sencilla
Fictional example: A clinician does not apply a general chatbot's quantity suggestion. The prescribed preparation and authoritative information are checked through the approved process, with medical or pharmaceutical escalation if needed.
Recomendaciones prácticas
Our recommendation: Separate writing assistance, research and treatment decisions in policy. Do not approve general assistants as independent sources for individual dosing. Define suitable systems and professional information sources.
For specialised products, review intended purpose, conformity documentation and limits. Test units, missing inputs and outdated information using fictional cases. Prompt restrictions are only one control layer; prevent unreviewed outputs entering actionable orders. Provide a clear escalation route for conflicting information.