How iDoctor stays safe.

Calm until it matters.

Color and alerts are saved for the moments that need them. The rules below hold in every one-to-one conversation with iDoctor, in Messages and on the web.

Emergencies come first

Before the AI reads a message, iDoctor checks it for emergency signs. If it finds one, it tells you to call 911, or 988 in a mental-health crisis, and stops there.

Your records stay yours

Download a copy of your records, or delete your account, whenever you like. iDoctor doesn't sell your data or show you ads.

Approvals and certifications will appear here when they exist, not before.

How physicians supervise the AIComing soon

In plain words: the AI drafts, physicians decide, and independence is earned one kind of task at a time.

  1. A physician approves, by name.

    New kinds of work go to a licensed physician before anything reaches you. When a physician writes to you, the message carries their name and title.

  2. Independence is earned in stages.

    A task moves from approved-first to spot-checked only after the AI agrees with physicians within a set margin of how often two physicians agree with each other, and never misses a case that must go to a person.

  3. Anyone can pull the brake.

    Any reviewing physician can escalate a case. Our medical director can suspend a task in one step, and it takes effect at once.

These always go to a physician

  • Chest pain or trouble breathing. You're told to call 911 first.
  • Pregnancy, or planning one.
  • A new medicine that could interact with yours.
  • Anything you ask a physician to see.
  • Any case where the AI isn't sure.
  • Every diagnosis. The AI never gives one on its own.

What went wrong

Every adverse event, in plain words, with what we changed. Serious events would be listed here first, within 72 hours.

How we countComing soon

Definitions behind every number on this page, so you can hold us to them.

Concordance
The share of physician-reviewed cases where the AI's proposed action matched the physician's final action. The lower bound is a Wilson 95% interval.
Escalation
A case the AI routed to a physician instead of drafting an action, because a rule required it or it wasn't sure.
Adverse event
Any error that reached a patient, whether or not it caused harm. Serious means it caused or could have caused harm.
Review time
Minutes from the patient's message to a physician's decision, counted only for cases that require one.