What security questions should a law firm ask a legal AI vendor?
The security questions a law firm should ask a legal AI vendor are: where client data is stored and processed, whether it is used to train AI models, how your data is kept separate from other customers', which certifications the vendor holds, who can access your data, and what happens to it when you leave. Get the answers in writing and check that the contract says the same thing as the website.
Most firms of 20 to 100 fee earners do not have a security team to run vendor due diligence. The partner sponsoring the tool, the person responsible for risk and compliance and whoever looks after IT usually share it. This checklist is meant to give them a common list, so every vendor on the shortlist answers the same questions.
What security questions should you ask a legal AI vendor?
Ask the eight questions below of every vendor, whether that is Harvey, Legora, Spellbook, Luminance, Microsoft Copilot or Scriptum AI. Comparing written answers side by side tells you far more than a sales demo.
| Question | Why it matters | A good answer looks like |
|---|---|---|
| Where is our data stored, and where is it processed? | Storage and processing can happen in different countries, and international transfers need a lawful basis | Named locations for both, with the transfer mechanism stated |
| Is our data used to train AI models? | Client documents should not improve a model other customers use | A clear no, covering the vendor and its AI providers, written into the contract |
| How is our data separated from other customers'? | Firms on both sides of a deal may use the same tool | Separation enforced in the system itself, not only by policy |
| Which certifications do you hold? | Clients and panels often ask for ISO 27001 or SOC 2 | Certificates you can see, or an honest account of progress |
| Who at the vendor can access our data? | Support access is a common gap | Need-to-know access, confidentiality obligations and logging |
| Who are your sub-processors? | Your data may pass through AI and hosting providers | A published, current list |
| Is our data encrypted? | A baseline expectation | Encryption in transit and at rest |
| What happens to our data when we leave? | Your precedents and clause history are firm assets | A defined export window and deletion timetable |
A starting checklist, not legal advice. Add your clients' and panels' own security requirements.
Why does it matter where legal AI processes data?
Because storage and processing are often in different places. A vendor can store documents in the UK or EU while sending text to an AI model provider elsewhere for processing.
That is not necessarily a problem, but you need to know it, check the transfer mechanism and reflect it in your data protection assessment. The ICO's guidance on AI and data protection is the place to start, and our post on GDPR and legal AI covers what firms usually need to do.
Should a law firm let a vendor train AI on its documents?
For client work, generally no. Your precedents and negotiated positions are confidential to your clients and valuable to your firm, and they should not improve a model that your competitors also use.
Check that the commitment covers the vendor's own AI providers as well as the vendor, and that it sits in the contract. The SRA's Risk Outlook report on the use of AI in the legal market is a useful read on confidentiality and AI more broadly, alongside our post on keeping client data confidential when using AI.
A vendor's security page is a promise. The contract is what you can rely on.
Do certifications like ISO 27001 and SOC 2 settle it?
They help, but they do not answer every question on the list. A certificate shows a vendor runs an audited security programme; it does not tell you where your documents are processed or whether they are used for training.
Younger vendors may still be working towards certification. That is worth knowing rather than a reason to stop, provided they say so honestly and answer your questionnaire in writing. The useful test is whether the vendor's answer matches what your clients and panels will accept.
What should an in-house legal team add?
Your company's own procurement and information security requirements. In-house teams rarely choose a vendor alone, so bring IT security in at the start rather than after the pilot.
The questions above still apply, and in-house teams should also check how the tool fits the company's Microsoft 365 set-up. Our guide to the best AI drafting tool for in-house legal teams covers the wider choice.
How does Scriptum AI answer these questions?
Scriptum AI publishes its answers on its security page. All customer data is stored in the EU, in Frankfurt, Germany. When an AI feature runs, document text is processed by AI providers in the United States but not stored there. Customer data is never used to train AI models, and each firm's data is separated at the database level, so firms on both sides of a deal cannot see each other's work.
Data is encrypted in transit and at rest, support access is logged, and every AI output needs a lawyer's approval before it takes effect. Scriptum AI is working towards ISO 27001 and SOC 2 but neither is complete yet, and we will complete security questionnaires for clients and panels on request. Your clause library and negotiation history belong to your firm, with a 30-day export window if you leave.
Running vendor due diligence? Book a demo and we will take you through our security answers alongside a live drafting matter.
Common questions
Is legal AI safe to use with client documents?
It can be, if the vendor stores data in a known location, does not train models on it, separates your data from other customers' and lets you delete it when you leave. Ask for those commitments in writing, in the contract, not only on a web page.
Should a legal AI vendor have ISO 27001 or SOC 2?
Independent certification is a strong signal, but it is not the only one. If a vendor is not yet certified, ask what stage it has reached and for written answers to your security questionnaire, and weigh that against your clients' and panels' requirements.
Do we need a data protection impact assessment for legal AI?
Often, yes. If the tool will process personal data in client documents, check whether UK GDPR requires a data protection impact assessment before a pilot on live matters. The ICO's guidance on AI and data protection explains when one is needed.
Do in-house legal teams need to ask different questions?
Mostly the same questions, plus your company's own information security and procurement requirements. In-house teams should involve IT security early, because they will usually have to approve the vendor anyway.
Where does Scriptum AI store client data?
Scriptum AI stores all customer data in the European Union, in Frankfurt, Germany. Document text is sent to AI providers in the United States for processing when an AI feature runs, but is not stored there, and it is never used to train AI models.