Trust
Privacy notice
Last updated 22 October 2026
Draft under legal review. Published so that customers and their advisers can read it before it is finalised; it may change before it takes effect.
This notice explains how Nansen Pty Ltd, trading as The Digital Safeguard Centre (“we”), handles personal information in connection with Harmwise Guardrail. It is written to meet the Australian Privacy Principles. Where we handle information on behalf of a customer, the customer's own privacy policy also applies, and our data processing terms govern what we do with it.
What we handle, and why
- Account information: the names and email addresses of people at our customers who use the console, their Microsoft tenant identifier where they sign in through it, and their roles and notification choices. When someone asks for a sign-in link by email, we keep their address and a hash of the link for a day, then delete them. We use it to give them access, to secure the service and to contact them about it.
- Billing information: the billing contact, address and ABN, and payment details, which our payment processor holds; we do not see card or bank numbers.
- Conversation content, handled for our customers: the reply their AI system sent and the turns before it, sent to us for grading. We do not keep the text of conversations. We keep the findings, signals and metadata, and a keyed hash of the conversation so the customer can check their own copy against our record. A request queued for grading is held only until it is graded. Quoted spans of findings are kept only for customers who choose to keep them, encrypted, and for the retention period they set.
- Technical information: request metadata (time, path, status, a request id) and error messages, used to run and secure the service. We do not log request bodies or query strings.
- Support requests: what a customer tells us when asking for help.
Conversation content may include sensitive information, including about family violence, health or crime, about people who never deal with us directly. We handle it only to provide the service to the customer who sent it, and we never use it to train a general-purpose AI model.
Who else sees it
The service providers listed on our sub-processor page, only for the purposes listed there. Our staff see customers' metadata and aggregate figures; access to retained quotes needs a stated reason and a second approver, lasts four hours, and appears in the customer's own audit log. We disclose information to authorities only where the law requires it, as our Terms describe.
Where it is held
The service's application, database and grader run in Australia (Sydney and Australia East). Some providers process limited information overseas: sign-in (Microsoft, GitHub, Google), payments (Stripe) and email (Resend), in the United States and elsewhere, as the sub-processor page says. Those providers handle account and billing information, not conversation content. One exception is chosen by the customer: an organisation that turns on the model judge has the turns it grades read by Anthropic, in the United States, and not kept by Harmwise.
How long we keep it
- Records of graded replies: seven years from when they were recorded, while the customer's account is open, then deleted after notice to the customer, unless a legal hold applies. After a year they move from the database to archive storage in Australia, unchanged. Retained quotes become unreadable after the customer's retention period (90 days by default).
- A closed account, and everything in it: deleted 28 days after closure.
- Queued grading requests: until graded, then the text is removed; the job itself after seven days.
- Billing records: as long as tax law requires.
Your choices and rights
You can ask for access to, or correction of, personal information we hold about you. If we hold it on behalf of one of our customers, we will refer you to them and help them respond. Notification emails can be switched off on the console's Team page. To ask, or to complain, write to us through the support page. If you are not satisfied with our answer, you can complain to the Office of the Australian Information Commissioner (oaic.gov.au).
A data breach
If a breach is likely to cause serious harm, we notify the people affected and the Information Commissioner as the Notifiable Data Breaches scheme requires, and we tell affected customers within 72 hours of becoming aware.