AI Policy
This policy explains how Scriptum uses AI and the principles we build to. It summarises commitments set out in our Terms of Service and Data Protection Policy, which govern if there is any difference.
1. A lawyer stays in control
Scriptum holds every AI proposal for a human decision. Agents prepare first drafts, review plans, negotiation responses and client questions. A named lawyer decides what is accepted, and nothing is released to the client or the other side without approval. Our negotiation agent works only inside limits a partner sets, and anything outside them goes back to the partner.
2. AI output can be wrong
AI output can be incomplete, inaccurate, out of date or unsuited to a matter, and can include invented or misattributed content. A suitably qualified person must review it and apply their own judgement before it is relied on, approved, released or sent to anyone. For more on professional duties, see the SRA's professional duties and using AI.
3. Scriptum does not give legal advice
Scriptum is a software provider, not a law firm. Using it does not create a solicitor and client relationship with us. Your firm remains responsible for the advice and documents it gives its clients.
4. We do not train AI on your data
We will not use your documents, or the clause library and negotiation history built from them, to train any AI model. We engage AI providers only on terms that do not permit them to train on it. Those terms may allow a provider to hold inputs for a limited period to detect misuse.
5. Your firm owns what it builds
Your clause library, playbook positions, negotiation history and counterparty records belong to your organisation, not to any individual user, and stay with your firm when people leave. You also own the output generated for you, to the extent rights can exist in it.
6. Cross-firm benchmarks are opt-in
Benchmarks across firms are off by default. They apply only if a firm opts in, and use only de-identified data presented so that no firm, matter or client can be reconstructed from it.
7. Which AI providers we use
We use AI models from OpenAI and Anthropic, processed in the United States and not stored there. The full list, with what each provider does, is on our sub-processors page. We may change the models and providers we use, subject to our Data Protection Policy, including at least 30 days' notice before adding or replacing a sub-processor.
8. Firms are kept apart
Each firm's data is separated from every other firm's at the database. Both sides of a deal may use Scriptum, and neither sees the other's data except what a firm chooses to release to a shared matter.
9. Questions and concerns
If you have a question about how Scriptum uses AI, or think something has gone wrong, contact us.
Common questions
Does Scriptum's AI send anything without a lawyer's approval?
No. AI output in Scriptum is a proposal. It only takes effect when an authorised user approves it, and nothing is released to the client or the other side without approval.
Which AI models does Scriptum use?
Scriptum uses AI models from OpenAI and Anthropic. Both are listed as sub-processors, with what each does and where processing happens.
Who owns the clause library Scriptum builds?
Your firm does. Your clause library, playbook positions, negotiation history and counterparty records belong to your organisation and stay with it when people leave.