Assistive, not autonomous
Sanctum Lex produces drafts, findings, and candidate answers for review by a qualified professional. It does not give legal advice and does not make decisions on a matter.
Legal / Responsible AI
Responsible AI statement
What the system does, what it does not do, and where a professional judgement is still required. Sovereignty solves confidentiality. It does not remove the need for a lawyer.
Grounded output
Answers are generated against documents in the enclave and returned with citations to the file and page. Where the corpus does not support an answer, the system is designed to say so rather than fill the gap.
Known limitations
Models can misread a clause, miss a document that retrieval did not rank, or state an inference with more confidence than the record supports. Verify anything that will be filed, sent, or relied upon.
Human review
Privilege designations, responsiveness calls, and anything reaching a client or a court should be reviewed by a person accountable for the matter.
No training on client content
Client documents, prompts, and outputs are never used to train, tune, or evaluate any model, inside or outside the enclave.
Evaluation
We evaluate retrieval and citation faithfulness before shipping model or runtime changes, and we share the methodology with clients under NDA.
Bias and fairness
Models reflect their training data. Where output touches individuals, treat it as one input among several and apply the same scrutiny you would to a junior colleague's memorandum.
Reporting a problem
Clients can raise output concerns through their named contact. We investigate using reproductions the client supplies, because we cannot see their content.
This page is a drafting starting point prepared for the site build. Have counsel review and adapt it, and insert the governing jurisdiction, before publication.
Questions about any of this?
The private office answers legal and compliance questions directly, and will put your reviewers in touch with the right person.