VARNIUMAI risk defense / everyday bearings

CONSENT / SYMPTOM BOUNDARY

AI Symptom Checker: When to Stop Trusting It and Call Someone.

A symptom checker can organize your symptoms and it can invent a diagnosis with equal confidence. The guide marks the exact line where the right move is a human, now.

Start here

The free tool below runs entirely in this browser — no account, nothing stored or transmitted. Redact anything you would not want kept before you type it.

Open the companion tool →

Why the boundary is the whole tool

A symptom checker is useful for organizing what you are feeling, and dangerous when it becomes a substitute for care. It cannot examine you, cannot order tests, and cannot be accountable for a missed emergency — and it will sound equally confident either way.

The boundary is a list, not a judgment: certain symptoms — chest pain, difficulty breathing, stroke signs, severe bleeding, suicidal thoughts — end the chatbot's usefulness. At that point, the next step is a human, and delay is the only wrong move.

The protocol

  1. List the symptoms you have asked about, without adding identifying detail.
  2. Read the boundary basis: which symptoms are safe to explore and which are not.
  3. For boundary symptoms, call a clinician or 911 now — do not keep asking the model.
  4. For everything else, use the model to organize questions for a real appointment.

What the tool does

Review symptoms you have asked about and get a boundary reading of which mean a human now. It runs in this browser; nothing is sent or stored.

Limit first: This is a boundary aid, not medical advice and not a diagnosis. It cannot examine you. For boundary symptoms, call a clinician or 911 now — delay is the only wrong move.

Open the companion tool →

One human next step

For boundary symptoms, call a clinician or 911 now. For the rest, bring the organized questions to a real appointment — a model cannot examine you.

Research log and safety checks

AI-assist path: not applicable — the tool's job is to mark the line where the reader must leave the model and reach a human; adding a model on top would be the exact failure it prevents. §6.10: not a frontier-pace topic.

§0.5 protection result: category is informed-consent-literacy; crisis-adjacent: no; fear/urgency/scarcity toolkit used: no. The protective function is free and the paid feature is convenience only.

Related dispatches

Pairs with the same protection bar: a free complete reading, an optional convenience unlock, and no fear-framing.