Get the app

Privacy

How LabSense uses AI

Which features use AI, what each one sends and to whom, how personal details are hidden first, and how to turn AI off.

Last updated

When we make a significant change to this page, we'll tell you in the app or by email before it takes effect. Smaller edits show up as a new date here.

In short

  • AI does the reading and the writing-up. It reads your lab reports, estimates what’s in your food, reads supplement labels and writes your Health Status.
  • It asks first. Nothing is sent to an AI service until you allow it in the app, and you can turn it off at any time.
  • It sees as little as possible. Each feature sends only what it needs. Names, birth dates and ID numbers on lab reports are hidden before the AI sees them, and your name, email and account ID are never sent.
  • It isn’t trained on your data. The AI providers process your data for LabSense and don’t use it to train their models.
  • It can be wrong. Check digitized values against your report, and treat everything as information, not a diagnosis.

Your choice

Before any AI feature first sends data, the app shows Allow AI processing:

LabSense uses AI to read your lab reports, analyze meals and supplement labels, and write your Health Status. To do that, it sends the data a feature needs to an AI provider.

Who gets it. Anthropic, which makes the Claude models. Voice descriptions of meals go to OpenAI to be turned into text.

What’s sent. Only what the feature needs: the report, photo, text or recording you add, and for Health Status your results and answers with just your sex and age. Names, birth dates and ID numbers on reports are hidden first, and your email is never sent.

How it’s used. They process it only for LabSense and don’t use it to train their models. Your records stay on LabSense’s servers in the EU.

Your choice. You can turn this off any time in Settings. Without it, you can still add results by hand and use everything that doesn’t need AI.

The switch is More → Settings → AI processing.

You can tap Allow or Not now. Your choice is saved with your account, with the date and the version of the text you saw. Our servers check it and refuse any AI work without it, so a setting on one phone can’t be bypassed from another.

If you turn AI off later, nothing more is sent. What was already analyzed stays in your account until you delete it.

What still works without AI: entering lab results by hand, biomarkers with their ranges, statuses and charts, body-system scores, checkups, supplement tracking and reminders, Apple Health and Health Connect sync, medical documents and the Wiki. Without AI, you can’t upload a report for automatic reading, log food by photo, description or voice, scan a supplement label or get a Health Status.

Who processes it

Provider What for Where
Anthropic, with its Claude models Reading lab reports, report titles, food analysis, ingredient and nutrient calculation, supplement label scanning, Health Status United States
OpenAI Turning voice descriptions of meals into text. Nothing else. United States

Both act as our service providers under their commercial API terms. Data leaves the EU under the safeguards described in the Privacy Policy.

Feature by feature

Reading a lab report

  • What’s sent: the redacted copy of a digital PDF, or the page images you painted over for a photo or scan. See Hiding personal details first.
  • What comes back: the report’s date and lab, each test’s printed name, value, unit and the lab’s printed range, and a title for the report. LabSense matches the tests to its catalog and saves the values to your profile.
  • What happens to the copy: it’s deleted from our storage once the analysis succeeds. Your original file stays only if Keep original files on our servers is on, and it’s never sent to the AI.

Titles for reports you enter by hand

  • What’s sent: the names of the biomarkers in the report, nothing else.
  • What comes back: a short title such as “Kidney panel”.

Logging food by photo or description

  • What’s sent: your food photo and the description you add, or just the description.
  • What comes back: the food’s name, its portion, its ingredients with their weights, and its nutrients.
  • The photo stays in your food diary for two weeks, then it’s deleted from our storage.

Describing a meal by voice

  • What’s sent: your recording, up to 60 seconds, goes to OpenAI with the language you chose (English or Russian).
  • What comes back: the text, which you can check and edit before it’s analyzed like any typed description.
  • The recording isn’t stored by LabSense. It’s passed on and discarded.

Editing ingredients and recalculating nutrients

  • What’s sent: the food’s name, its amount and its ingredient list.
  • What comes back: updated ingredient weights or nutrients.

Scanning a supplement label (LabSense Plus)

  • What’s sent: the photo of the label.
  • What comes back: the nutrients, their forms and amounts, and the other ingredients.
  • The photo isn’t stored. It’s read and discarded.

Health Status

  • What’s sent: the profile’s sex and age, its results from the last 6 months with each value’s status and normal range (at most the 12 newest per biomarker), and its survey answers: conditions, symptoms and lifestyle. For the account holder’s own profile, also the supplements taken (their nutrients, not the names of supplements you created), how regularly they were taken, the daily average of the last 30 days of food, and daily steps and sleep imported from Apple Health or Health Connect.
  • What comes back: a short summary, possible problems and future risks to look into, and recommendations.
  • What it doesn’t do: the body-system scores aren’t written by AI. They come from a fixed formula over your values and answers.

Hiding personal details first

Lab reports usually carry your name, date of birth and ID numbers. The AI doesn’t need them, so they’re hidden before it reads anything.

Digital PDFs: redacted on our servers

When you upload a PDF that has a text layer, our own redaction service, running on our servers in Frankfurt, finds and removes personal details:

  • labeled fields such as name, date of birth, address, phone and policy numbers, in English and Russian
  • phone numbers, email addresses, national ID and insurance numbers, and postal addresses, by their shape
  • lab order, sample, patient and medical record numbers, which with some labs open your full results online
  • names that aren’t labeled, found by a language model that runs on our servers, not an AI service
  • the details LabSense already knows: the profile names and email on your account, and a full date in a profile’s birth year

It removes the text underneath, not just covers it, and also removes every image (logos, stamps, signatures, QR codes), the document’s metadata, links, attachments and earlier versions. Then it reads the result back to check that nothing it hid is still there. Dates of the test, the lab’s name, sex, age and every result are kept, because they’re needed to read your results correctly.

You see the redacted pages before anything is sent, with a list of what was hidden. You can hide more, or un-hide something hidden by mistake, then tap Start analysis. If the check fails, or the PDF is really a scanned picture, you’re asked to paint over the details yourself instead.

The redaction service is open source, so anyone can see how it works: github.com/Sultannis/labsense-redactor.

Photos and scans: you paint over them

For photos, document scans and scanned PDFs, you paint over your personal details in the app before sending. The painting is baked into the images on your phone, so the details never reach the AI. Anything you leave visible will be read by the AI, and the app reminds you of that. If Keep original files on our servers is on, the unpainted original is stored for you too, but it’s never sent to the AI.

What’s never sent

  • your name, email address, account ID or exact date of birth
  • profile photos
  • medical documents such as ultrasound, ECG or MRI reports
  • your sign-in details, device identifiers or location

What the providers do with it

  • They process it for us. Anthropic and OpenAI act as our service providers and use your data only to return the result to LabSense.
  • They don’t train on it. Under their API terms, they don’t use it to train their models.
  • They may keep it briefly. They may keep inputs and outputs for a limited period to detect and prevent abuse, and then delete them. We’ve asked both for zero data retention, which would end this, and we’ll update this page if they grant it.

What isn’t AI

These parts of LabSense follow fixed rules and reviewed reference data, not AI:

  • reference and optimal ranges, and whether a value is in range
  • body-system scores
  • which tests your checkup lists
  • nutrient targets and supplement upper-limit warnings
  • the Wiki

Accuracy and limits

  • Digitized values can be wrong. A value, unit or match to a test can be misread. LabSense marks values the AI is unsure of with a ? for you to check, and you can correct any value on the report’s page. Some tests aren’t supported yet; the report lists them instead of guessing.
  • Food estimates are estimates. Portions and nutrients from a photo or description are approximate. You can edit the portion and ingredients.
  • The Health Status isn’t a diagnosis. It points out patterns worth discussing with a doctor. It can miss things or flag things that turn out to be fine.
  • AI can decline. The AI provider’s safety checks occasionally refuse a request. If that happens, the analysis fails and any free use is given back.
  • Daily limits apply. Every account, Plus included, has a generous daily limit on AI features, to protect the service from misuse.

LabSense is for information only and isn’t a medical device. Read the Medical disclaimer.

Questions

Write to privacy@labsenseapp.com. The Privacy Policy has the full details.