Safety

A macro and activity tracking companion. Not a medical device. Not weight-loss treatment. Not a substitute for a doctor, dietitian, or therapist.

At signup: a brief 5-question screen (SCOFF) for eating-disorder indicators. Anyone scoring in the higher-risk range is not onboarded and is referred to specialist resources instead.

On every voice note: a safety classifier looks for three patterns — subclinical-ED language (6 sub-categories), out-of-scope requests (medication, medical conditions), and acute crisis. The coach responds differently depending on what fires.

• Voice note audio: 30 days
• Voice note transcripts: 1 year
• Structured food/exercise logs and summaries: indefinite (until you delete)
• Audit log of safety events: indefinite (anonymised when you delete your account)

You can request a download of everything or hard-delete your account at any time. Both happen within 30 days.

• Streaks that punish you for missing a day
• Leaderboards or comparison to other users
• Public weighing or before/after imagery
• Outcome-based pricing tied to weight loss
• "Good/bad" food labels or "cheat days"
• Carry-over calorie deficits between days
• Net-calories-after-exercise as a primary metric
• Photo body-progress features
• Selling or recommending supplements
• Sharing your data with advertisers — ever

If our system detects acute distress, we send a static message containing helpline numbers for your country (Beat, Samaritans, Crisis Text Line, 988, ANAD, iCall, Vandrevala, etc.). We do not generate that message with AI — it's a fixed template with hardcoded numbers.

We pause coaching for 14 days after a crisis-tag fires, so you can use the support channels without us in the way. You can resume anytime.

Email safety@dietfam.app. We acknowledge within 24 hours and aim to resolve within 7 days. Material incidents are posted to dietfam.app/safety/incidents.

Every week, we run an adversarial AI review on 20 random transcripts (anonymised) to look for sycophancy, identity-praise, fabricated nutrition claims, missed ED signals, and other failure patterns. Every month, a different AI model audits those reviews for drift. Findings are filed under decisions/safety-reviews/ in our repo as a public audit trail.

When a serious finding emerges, we consult a credentialed clinician ad-hoc (we don't have a retained advisory board at this stage).

Companies that handle your data on our behalf: AWS, Meta (WhatsApp), Anthropic (Claude), Google (Gemini for speech-to-text), Stripe, Langfuse. Full list and locations at dietfam.app/sub-processors. We notify you 30 days before adding or changing any sub-processor.

Privacy: privacy@dietfam.app · Safety: safety@dietfam.app · General: hello@dietfam.app