---
title: "Security"
description: "How SafetyLM keeps your questions, documents and chats private, and the checks it computes outside the model."
url: https://safetylm.ai/security
---

SECURITY Built in, not bolted on.

# Your questions stay yours.

SafetyLM is built so what you ask, attach and save stays private, and so the checks that matter are computed by the system, not requested of a model.

[Read the privacy policy](https://safetylm.ai/privacy)

## Our commitments

### Stored Locally in Australia

Your data is kept in your account in Australia, behind row-level security.

### Never Used to Train Our AI Models

Your questions, documents and chats are never used to train our AI models, and we never sell them.

### Documents Stay With You

An attached document is read in your browser and sent only with your question. It is never stored.

### Declined, Never Estimated

If the law, the guidance or a figure can't be backed by its sources, SafetyLM declines, often before the model even runs.

### Australian Privacy Principles

We apply them whether or not the Privacy Act requires it.

## What happens to what you ask.

### Where your data lives

- **Stored locally in Australia**: Your data is kept in your account, in an Australian data centre in Sydney, behind row-level security.
- **Answered on Cloudflare's network**: Questions are answered by models running on Cloudflare's global network.
- **Web search through Anthropic**: When an answer needs the web, the search goes through Anthropic's API. It never searches for a penalty figure.

### Checks that are computed

- **The refusal**: If the law, the guidance or a figure can't be backed by its sources, SafetyLM declines rather than estimate. A penalty question with no verified figure is declined before any model runs.
- **The citation check**: Every cited section is matched against the retrieved passages and marked found or not found.
- **A second pass**: Each regular answer is read again before you see it.
- **Logs without content**: Every turn is logged with the software version, never the question or the answer.

### Your account, your control

- **By invitation**: Accounts are opened by invitation, with email and password sign-in.
- **Delete what you want**: Delete a chat, or your whole account, from Settings.
- **Card details never reach us**: Payments are handled by Stripe. Card details never reach SafetyLM.

What we don't claim yet. We don't hold SOC 2 or ISO certifications. When we do, they'll be listed here with their reports.

## See how SafetyLM fits the way you work.

SafetyLM is opening in small groups. Join the waitlist and we’ll email you when your place comes up.

[Join the waitlist](https://safetylm.ai/waitlist)