Safety

AI limitations

Last updated: 21 August 2026

How Alex uses bounded context, where generative AI can fail and which Gymanai rules remain authoritative.

1. What Alex is and is not

Alex is a bounded server capability for fitness explanations and validated product actions. It does not diagnose, treat or monitor health conditions and is not a clinician, emergency service, unrestricted agent or independent authority over your account, plan, billing or safety state.

2. Generative output can be wrong

AI output can be incomplete, outdated, unsuitable or wrong. It can misunderstand a question, omit context, invent details, use stale knowledge or express uncertainty poorly. Fluent language is not proof of accuracy.

3. Context, language and memory

Gymanai sends bounded context and tells the model to answer directly in the active language; it does not make a second translation-model call. Health-sensitive facts require separate current consent before activation in coach memory. Memory is server-owned and member-controlled.

4. Deterministic authority and actions

Fitness-safety, training and account-access rules take priority over AI output. AI may explain or propose, but it cannot bypass ownership checks, quotas, schemas, current-state validation or required confirmation.

5. Safety and reporting concerns

Do not use Alex for diagnosis, treatment or emergencies. Stop exercise and seek appropriate local help for concerning symptoms. Use Report this response for unsafe or clearly incorrect output. Contact gymanai.team@gmail.com; a verified support route is also still a pre-launch requirement.

Questions about these pages can be sent to gymanai.team@gmail.com.