Safety and privacy

Safety, privacy and clinical governance

Vet Amica is a communication tool, not a clinical decision-maker. This page explains how the vet stays in control, which parts use AI, the checks that help catch mistakes, and what happens to a practice’s reports and data.

Last updated 1 October 2026.

The vet is responsible, as for any letter

The vet who approves an owner sheet is responsible for it, exactly as for any letter or report they send. That vet knows the animal, the case and the owner.

Approval cannot be skipped. No setting turns it off, and nothing can be shown to the owner before it.
The vet confirms they have read it, and approves under their own name, chosen from the practice’s list of vets.
Everything is editable. The vet can rewrite any line, remove any box, and hide or show individual results.

Some parts use AI. Others use none at all.

The written summaries use AI (Claude, made by Anthropic). Everything else is fixed.

Blood and urine values: No AI. Each value is shown exactly as the laboratory printed it, beside a plain-English name and description from a fixed list of about 140 tests.
“What was checked” and “What the numbers showed”: No AI. Filled in from the values; they only count results and repeat the laboratory’s own flags.
The other answers: Drafted by AI from the laboratory report and the vet’s case notes, then edited by the vet.
Imaging answers: Drafted by AI from the vet’s own written findings. X-ray and ultrasound pictures are never sent to the AI.
Estimate of costs: Taken from the vet’s own figures or the practice’s price list. Never invented.
Condition information sheets: Written by a practising vet, not by AI.

Checks that help the vet catch what’s easy to miss

These appear only on the vet’s screen, never on the owner’s sheet or in print. They never replace the vet’s judgement.

Numbers

The tool never decides for itself whether a blood value is high or low, and never invents a reference range. It repeats the laboratory’s own flags.
Every value is checked against the range printed beside it. If a flag is missing or doesn’t match, the vet is warned, because that points to a misread.
Lines the tool doesn’t recognise are flagged for the vet, never guessed.
A report read from a photo, scan or screenshot carries an extra “please check these figures” warning.

Medicines

The AI never suggests a medicine or a dose of its own. It can only repeat one the vet has written.
Any medicine named in what the owner will read must be confirmed by the vet before approval unlocks.
Fifteen built-in checks highlight combinations worth a second look, such as an anti-inflammatory with a steroid. They are worded “worth a look”, never “this is wrong”.

Prices

Every figure comes from the vet’s notes or the practice’s price list, and the vet’s own figure always wins.
A cost line nobody asked for is flagged, and so is a priced step that didn’t reach the estimate. Estimates are labelled as a guide, not a quotation.

Wording

The AI is told to describe only what is in the laboratory report or the vet’s notes, and never to add a finding of its own.
Where the laboratory was uncertain, the AI is told to keep the owner’s version uncertain. The vet’s read is the check on this.

Where the vet is the safeguard

General background is not independently fact-checked. Where the AI explains a condition in general terms, the approving vet is the safeguard.
The tool only knows what it is given: the report and the vet’s notes, not the whole clinical history.
Photos can be misread. Decimal points and HIGH or LOW marks are easily lost in a picture.
Dogs and cats only. The descriptions, condition sheets and medicine checks are written for these two species.

Keeping the clinical content up to date

The test descriptions, the instructions the AI follows, the medicine checks and the condition sheets are reviewed frequently by a practising vet. Practices are told when clinical content changes.

A “Report a problem” button in the tool lets a vet tell us the moment something looks wrong. Anything that could affect what an owner is told is dealt with first.

What happens to a practice’s reports and data

When a practice uses Vet Amica, the practice decides what goes in and is responsible for it (the “data controller”). We act on the practice’s instructions, under a written agreement signed before any real patient information is handled.

Where the data goes

To draft a summary, the report text and the vet’s case notes are sent, encrypted, through our web host (Netlify) to the AI service (Anthropic). Both companies are based in the United States.
A report added as a photo or screenshot is sent the same way, to be read.
X-ray and ultrasound pictures never leave the vet’s computer.
The full list of companies that handle practice data for us is below.

The companies that handle practice data for us (sub-processors)

To run Vet Amica we use the companies below. Each one handles some of a practice’s information on our behalf. Data protection law calls them “sub-processors”, and this is the list our agreement with each practice refers to.

Anthropic (United States): The AI service, Claude. It drafts the summaries and reads reports added as a photo or screenshot. It receives the report, the vet’s case notes and the pet’s name.
Netlify (United States): Our web host. It runs the tool and the practice sign-in, and carries each request to the AI service without storing it. It holds the practice’s details and any “Report a problem” messages.
Submarine (Guernsey): The developers who build and support the tool for us. They can see system settings and “Report a problem” messages when fixing a fault.
Google Workspace (United States): Our email. It holds only what a practice chooses to email us.

Anthropic, Netlify and Google are in the United States, so information sent to them leaves the Bailiwick of Guernsey and the UK. These transfers are made under the safeguards the law provides for.

Before we add or replace a company on this list, we will email each practice and give it the chance to object, as our agreement with the practice sets out.

This list was last updated on 1 October 2026.

What is kept, and for how long

At the AI service: Anthropic deletes what it receives within 30 days, unless it must keep something longer to prevent misuse or by law. It does not use the content to train its AI.
Unfinished work: Saved only in the vet’s own web browser, on their own computer, so a draft isn’t lost if the page is closed. It is discarded after 12 hours.
“Report a problem” messages: Kept by our web host for 12 months, then deleted.

Who can see it

The tool is reached through the practice’s own sign-in, and each approval is made under a named vet.

Security

Everything travels encrypted between the vet’s computer, our web host and the AI service.
The key that connects us to the AI service is held only in our web host’s secure settings, never in the website’s files.

When a practice leaves

We remove the practice’s sign-in and delete its price list and abbreviations list.

If something goes wrong

If we ever find a breach affecting a practice’s data, we will tell the practice within 72 hours of finding out, and work with them on what happens next.

The law we work to

Vet Amica Limited is registered in Guernsey and works to The Data Protection (Bailiwick of Guernsey) Law, 2017, and the UK General Data Protection Regulation. Our privacy notice covers the website and emails to us.

Questions?

Email hello@vetamica.co.uk to book a video call with the vet who built Vet Amica.

← Back to the main page