A conversation about building a safe AI agent, from the 2026 International GM1 Virtual Community Conference
At the 2026 International GM1 Virtual Community Conference, Mohsen Hejrati, co-founder and CEO of Basion, shared why artificial intelligence designed for people with rare disease must be built differently from the tools most of us encounter online. His central message was simple: for families navigating GM1 gangliosidosis, safety, context, consent, and community knowledge are not optional features. They are the starting point.
Families affected by GM1 gangliosidosis become experts in a kind of complexity that most health systems were not designed to manage. There may be appointments across neurology, genetics, nutrition, rehabilitation, primary care, gastroenterology, and other specialties. There can be years of laboratory results, imaging reports, medication histories, therapy notes, school records, and caregiver observations.
At the same time, families are trying to understand research, make decisions, coordinate care, and live their lives.
Artificial intelligence may be able to help with some of this work. But only if its designers keep the real lives of patients and caregivers in mind.
Why generic AI is not enough
Most medical research and many digital tools center on the “average” patient. Rare diseases challenge that approach. GM1 is not average, and neither is any individual living with it.
A generic AI tool may be able to summarize a webpage or answer a broad question. But it may not know the person asking the question. It may not have access to, or be able to meaningfully use, a long and complex medical history. Without that context, even an answer that sounds reasonable can be incomplete or inappropriate for a particular family.
As Hejrati explained during the conference, a safe AI agent for rare disease has to address three connected challenges:
- Understand the person, not just the question. A question about a medication, symptom, feeding issue, or possible research opportunity cannot always be answered safely in isolation. The answer may depend on diagnoses, allergies, current medicines, prior test results, and the experiences that shape a family’s day-to-day care.
- Stay grounded in the full medical and research landscape. Rare-disease knowledge is often scattered across scientific papers, clinical-trial registries, specialist guidance, and small but important pieces of evidence. A simple web search may surface only the most visible results, not necessarily the information most relevant to a rare condition or an individual situation.
- Be tested for the situations families actually face. Developers must carefully evaluate tools intended to support people with complex medical needs, with clear safety guardrails. It is not enough for an AI system to perform well on general questions. It must recognize uncertainty, avoid overstepping, and respond appropriately when the stakes are high.
The knowledge families carry matters
Medical papers and clinical guidelines are essential, but they are not the whole story.
Families living with GM1 develop practical knowledge that is often difficult to find anywhere else: what made a care transition easier, what questions helped during a hospital visit, how a feeding routine worked at home, or what support made school participation more possible. That lived experience does not replace clinical advice. But it can be profoundly valuable to another family facing a similar challenge.
During the presentation, Hejrati described how an AI agent can recognize when a caregiver’s firsthand experience may be useful to others and ask permission before sharing it. That detail matters. A responsible system should not treat personal stories as material to collect. It should give families control over whether, how, and when they share their experience.
This is one way technology can strengthen, rather than replace, community. With consent, it can help families find relevant perspectives and reduce the isolation that so often comes with a rare diagnosis.
A safe AI agent builds in privacy and consent
Health information is deeply personal. For rare-disease families, it can include not only medical records but also genetic information, caregiving experiences, and details that affect everyday life.
Basion’s presentation emphasized a commitment to patient ownership of data, no sale of patient data, and consent as a core part of the platform’s design. These principles should be central to any technology intended to support patients and caregivers.
Families deserve clear answers to basic questions:
- Who can see my information?
- What is it being used for?
- Can I choose whether to share it?
- Can I change my mind?
- Will my data be sold or used in ways I did not agree to?
No one can add trust at the end of a product-development process. It has to be part of the experience from the beginning.
A safe AI agent should make the human work lighter
The goal of a safe AI agent is not to replace physicians, caregivers, or the GM1 community. It is to make some of the overwhelming work more manageable.
For families, that could mean help organizing medical information, preparing for appointments, understanding research in plain language, or identifying questions to bring to a clinical team. For caregivers, it may mean having a clearer picture of the many moving pieces involved in care. For the broader community, it may mean that practical knowledge can travel farther, while each person remains in control of their own information.
The most important measure of success is not whether an AI tool seems impressive. It is whether it is genuinely useful, respectful, transparent, and safe for the people who rely on it.
Building with the GM1 community
Cure GM1 Foundation is grateful to the families, caregivers, advocates, and researchers who are helping shape more thoughtful approaches to technology in rare disease. Tools for the GM1 community should reflect the knowledge of the people who understand its needs best: those living with GM1 and those who care for them.
As these tools continue to develop, community feedback will remain essential. Families should be able to ask hard questions, set boundaries, and help define what useful support looks like.
The future of rare-disease technology should not be about asking families to adapt to another complicated system. It should be about building systems that adapt to families, respect their choices, and help them carry less alone.
Watch the conference presentation: “Building a Safe AI Agent for GM1”
This article reflects a presentation at the 2026 International GM1 Virtual Community Conference. AI tools do not replace medical care. Families should continue to discuss medical decisions and urgent concerns with their clinical team.