Legal
AI processing consent
- Version
draft-2026-11-20- Dated
- 20 November 2026
- In-app consent version
ai-processing-2026-11-draft-3
Draft — being finalised with our lawyer
Notes for counsel are in block quotes. The app asks for this consent before any AI request is sent. If the person says no, nothing is sent. The answer is stored with the version, the time, and where they agreed (next to a feature, or in Account settings). Withdrawals are kept as history. Changing the wording means a new version, and everyone is asked again.
Part A — Short text for the in-app dialog
Code: AI_CONSENT_TEXT in packages/validation/src/consents.ts. It has three fields:
- title: about 5 words;
- body: about 120 words; and
- draftNote: to be removed or emptied once final.
Title: Allow AI processing?
Body:
To use AI features, Fita Link sends what you give it to our AI provider, Anthropic, in the United States.
- Meal scan: your photo.
- Food estimate: the food you typed.
With both of these, and with plan notes, we send your goal and daily targets. The AI Trainer also gets this week's workouts, your food totals for the last 7 days, your latest weight and BMI, and yes/no answers from your health check.
Photos aren't kept. Chats stay in your account until you delete them, and are never shared with a gym. Results are estimates, not medical advice.
You can withdraw at any time in Account → AI processing.
Buttons: "Allow" / "Not now"
[LAWYER: (1) Whether this one consent also covers the cross-border transfer to the USA, or a separate statement is needed. (2) The previous in-app text (
ai-processing-2026-11-draft-2) said "The provider doesn't use any of it to train its models". It was removed indraft-3(decision S19-04) and is left out until confirmed against Anthropic's current commercial terms and the DPA. Add it back only if confirmed. (3) Whether a link to Part B and the Privacy Policy is needed in the dialog. A link is recommended.]
Part B — Full explanation ("Learn more" / Privacy Policy annex)
B1. Which features use AI
| Feature | What it does |
|---|---|
| Meal photo scan | Estimates the foods and portions on your plate from a photo |
| Food estimate | Estimates the nutrition of a food you type in |
| Plan notes | Writes short notes about your guided plan when you ask |
| AI Trainer | Answers your questions about your training, food and plan in a chat |
The AI provider is Anthropic (the Claude models), reached through its API. [CONFIRM: processing location and the provider's data retention under our agreement.]
B2. What is sent to the AI provider
| Feature | What is sent |
|---|---|
| Meal photo scan | The photo (shrunk on your device first), which meal it is, and (if you have a plan) your goal, daily targets, the calories planned for that meal and the calories left today |
| Food estimate | The food name and portion you typed, and (if you have a plan) your goal, daily targets and the size of a main meal in calories |
| Plan notes | Your goal and timeline, fitness level, where and when you train, days per week, this week's workouts, your daily calorie and nutrient targets, and whether the plan is in its cautious mode |
| AI Trainer | Your message and up to the last 20 messages of that chat. Also: your plan goal, timeline and week; height, weight and goal weight; targets and maintenance calories; training days, place and level; this week's workouts (done, missed, coming up); daily food totals for the last 7 days; latest weight and BMI. Your health safety answers go as yes/no only (injury, pregnancy, heart condition or high blood pressure, other condition or medicine) |
(Checked against the prompt builders in apps/api/src/coach/coach-prompts.ts and apps/api/src/ai-trainer/trainer-context.ts on 2026-11-20. Re-check if those files change.)
B3. What is never sent
- The optional free-text note you write in the health safety screen.
- Your name, email address, phone number or date of birth. [CONFIRM: no identifiers are included in prompts.]
- Data about any other person, including other members of your gym.
- Anything at all, if you have not given (or have withdrawn) this consent.
B4. What is stored
- Photos: not stored by Fita Link. A photo is used for the scan and then discarded. The food entries you check and save are kept in your food diary.
- Food estimates and plan notes: the results you save are kept with your diary or plan.
- AI Trainer chats: kept in your account, visible only to you, until you delete a chat or your account. [PLACEHOLDER — maximum retention if decided, e.g. 12 months.] They are included in "Download my data". They are never shared with a gym and cannot be shared.
- Usage records: we count AI uses and their cost per feature, without the content, to run daily limits and a monthly budget.
- At the AI provider: [CONFIRM: retention period and whether data is used for training, per Anthropic's terms and our DPA.]
B5. Limits of AI
- AI results are estimates and can be wrong. Photos can be misread and portions misjudged. Check everything and edit before you save.
- Foods we recognise use values from the FAO/INFOODS West African Food Composition Table (2019). Other foods are labelled "AI estimate".
- The AI Trainer follows safety rules. It does not diagnose, does not advise on medicine or supplement doses, and tells you to see a doctor about symptoms, injuries, pregnancy or medical conditions. It never suggests eating below your plan's calories or losing weight faster than about 1% of your body weight a week. It can still make mistakes.
- This is not medical advice. Do not use AI features in an emergency.
B6. Using AI through your gym
If your gym gives you an AI licence, you still have to give this consent yourself. The gym never sees your chats, photos or what you asked.
B7. Withdrawing consent
- Go to Account → AI processing and withdraw. You can give consent again later.
- What happens: no further data is sent to the AI provider. AI features stop working for you until you consent again. Everything else in Fita Link keeps working.
- What is not undone: data already sent before you withdrew. Your saved chats stay in your account until you delete them. [LAWYER: should withdrawing also offer to delete chats?]
B8. If the wording changes
If we change what is sent or how it is used, we will change this text and its version, and ask you again before the next AI use.
