Security is non-negotiable.
Your contracts are among the most sensitive documents you hold. This page states exactly how they're protected, who processes them, and what happens if something goes wrong. No vague promises, and no separate page for the parts that are harder to answer.
Contract data is never used for AI model training. Uploaded contracts and review outputs are processed solely to fulfil your review request, and both model providers we call operate under API data processing addendums that contractually prohibit training on customer data.
One document, start to finish.
Six stages, from the moment a contract reaches us to the moment it stops existing. Every claim below is answered in more detail further down this page.
Over TLS 1.3, from the app, the add-in, a forwarded email or an API key.
Bound to your organisation before anything reads it. Unrecognised senders are rejected, not quarantined.
Under contracts with our model providers that give no training rights and no retention.
Encrypted at rest in the UK and EU, in storage no other tenant can address.
Attributed to a person or a key, exportable by your admins.
On your retention schedule, or the day you leave, documents and object storage both.
Answered without the hedging.
If your form needs these in its own wording, send it over, we fill in questionnaires ourselves rather than making you chase.
No. Not us, and not the model providers we call, that’s a contractual term with them, not a setting. Nothing you send becomes training data for anyone, and we don’t keep your clause text to improve the product.
Only with a reason recorded, for a limited window, and it lands in the audit log you can export. It happens when you ask us for help with a specific review. Nobody browses customer contracts, and there is no path to do it quietly.
The UK and EU, on named sub-processors listed below. If a change would move your data outside that, you hear about it before it happens rather than in a changelog afterwards.
Row-level security in the database, so isolation doesn’t depend on application code remembering to filter. Quotas, feature gates and API scopes are enforced server-side on the same basis, including for the assistant connector.
The reviewer assigned to that one deal, and only for as long as it’s with them. They’re under confidentiality terms with us, their access ends when the revision is released, and it’s all in the log.
Your admins are told within two hours of us confirming an incident that affects your data, what we know, what we don’t yet, and what we’re doing. A full report with root cause and impact follows within 72 hours, and the ICO is notified where required.
Who can see what, exactly.
Four kinds of people touch a review, and only one of them is us.
"On request" is the honest version of support access: you ask for help, an engineer gets a time-boxed grant with the reason attached, and the grant itself appears in your audit log. Approved senders never get an account, they forward a contract from an address you've allowed and get the review back.
What actually happens to a document you upload.
The analysis calls a model provider. That is the part people want to understand properly, so here it is in plain terms rather than buried in a policy.
Your data goes in, your report comes out, and nothing is retained for training. That is a contractual term with our providers rather than a setting we could quietly change.
AI providers process data via EU-based endpoints, and sub-processor agreements include Standard Contractual Clauses where applicable.
Outputs are review indicators to assist a human decision. They are not automated decisions with legal effect, and you can ask for human review at any point.
Every third party that touches your data.
Listed here rather than on request, with 30 days' notice before any change.
Legal entities and the current agreements are annexed to the DPA. Adding or replacing any of these is notified in writing 30 days beforehand.
What we can put in front of you.
All of it exists today and you can have it on the call rather than after a procurement cycle. Ask for a certificate, a completed questionnaire or the sub-processor list and it comes back the same week.
Signed as part of onboarding, with sub-processors named and change notice in writing.
Send us yours and we complete it. A founder answers it, so the replies are specific.
Your admins pull the full log themselves, including on the way out. No ticket, no waiting on us.
Vern runs on infrastructure holding SOC 2 Type II and ISO 27001, audited annually. Certificates and the current scope are available on request.
Offboarding purges your documents and object storage in full, not a soft delete with a flag on it. You take audit exports and your playbooks on the way out, and we keep only what we're legally required to keep for invoicing.
Your data rights, and how to use them.
Ask for any of these in one line to the address below. We confirm in writing what we did, within statutory time.
If something goes wrong, this is the order it happens in.
The incident is detected and contained, and the internal response team is activated.
Affected customers are notified with the initial detail and the scope as we understand it.
Continuous updates while the investigation runs and remediation is applied, rather than silence until it is tidy.
Root cause, impact assessment and preventive measures, with the ICO notified where required.
Bring your security questionnaire to the call.
We'd rather answer it live than send you a PDF that dodges the hard rows. Ask us the awkward ones.