Safety review for patient-facing chat

Every reply to a patient, checked against your own protocol.

SafeRounds reviews the chats your AI assistant and support staff have with patients. It flags replies that put people at risk, shows the exact words behind each flag, and sends them to a person.

Request a free audit See it review a chat ↓

Try a conversation:

1The chat, as the patient saw it

Harbourline SupportOnline

2SafeRounds' review

Synthetic conversations from Harbourline Support, a fictional service. The verdicts come from a real run of SafeRounds on Claude, with its explanations shortened.

Five checks

Each reply is judged against your rules, not a generic idea of safe.

You bring your protocol, reference facts and patient-record fields. SafeRounds holds every reply to them.

Safety

Advice that could cause harm if followed, like changing a dose without the prescriber.

Escalation

A red flag not sent to the right help, plainly and in the same reply: 000, same-day, or routine.

Accuracy

Anything that contradicts your reference sheet, including your own numbers and hours.

Scope

Staff working beyond their role, like a support agent giving clinical advice.

Grounding

Replies that contradict or ignore the patient's record, or invent details.

How it works

From transcript to reviewer.

Send your chats

Through the API, or as a CSV or helpdesk export. Names, phone numbers, emails, Medicare numbers and addresses are removed before anything is stored or reviewed.

Each reply is reviewed

Claude reads the chat with your protocol and the patient's record, and quotes the exact words behind each verdict. Anything it can't verify goes to a person. Nothing passes by default.

People decide

Your reviewers confirm or overturn each flag. Their corrections become test cases, so the review keeps improving for your service.

Once running, SafeRounds alerts you when flag rates drift, and blocks any prompt or model change that catches fewer serious problems.

Evidence so far

What has been tested, and what hasn't.

What	Result	Status
First run on the real Claude API	16 of 16 verdicts matched the planted answers, across 3 synthetic conversations.	Done
Cost per conversation	About US$0.015 each, or US$0.007 when reviewed in bulk.	Measured
Labelled validation	250 conversations with planted failures and tricky near-misses, scored against human labels and against how often two physicians agree.	In progress
Design partners	None yet. Looking for two or three Australian services to try a free audit.	Open

These are early results on synthetic data. SafeRounds is a working prototype by a solo founder, not yet a validated product. Validation results will be posted here.

Patient data

Built for patient conversations from the start.

De-identified firstPersonal details are removed before a chat is stored or reviewed.
Hosted in AustraliaThe service and database are being set up in Sydney. Reviews run on Anthropic's API.
Every read loggedViewing a chat writes an audit record. Logs never contain chat text.
Not clinical adviceSafeRounds flags chats for human review. It doesn't diagnose, treat or replace a crisis service.

See what SafeRounds finds in your chats.

Send 20 de-identified transcripts and your protocol. You'll get back every flagged reply, with the evidence, at no cost.

Request a free audit

Or email [email protected]