AI Use and Limitations
Last updated 2026-10-13 · Effective from not yet in force
Draft statement describing the intended use of AI in LexSalim. It is not an approved policy and must be reviewed alongside the Terms and the Privacy Policy.
01What the AI does
LexSalim uses large language models and document-processing models to extract text from documents, summarise them, extract facts such as dates, parties, amounts and obligations, answer questions about the documents in a case, and draft notes, letters and schedules. Comparison of document versions and research across approved sources are planned and are labelled as such in the Service. These are assistance functions: they work from the material you provide and, where a jurisdiction pack is available, from a curated library of approved sources.
02What the AI does not do
It does not give legal, tax, accounting or financial advice. It does not represent you, file anything, send correspondence or make decisions. It does not guarantee accuracy or completeness. It can misread documents, miss context, produce confident statements that are not supported by the sources (sometimes called hallucination), or apply the wrong jurisdiction’s rules. It has no knowledge of your matter beyond the case it is working in.
03Human review
Every finding and draft is marked as proposed until a person reviews and confirms it. Findings carry references to the pages or sources they rely on so they can be checked against the original. You are responsible for reviewing output before relying on it, and for involving a qualified professional where the matter calls for one. Deadlines suggested by the AI are proposals and must be verified against the applicable rules.
04Model providers and your data
Analysis uses third-party model providers as sub-processors under contracts that prohibit training on Customer Content and limit retention. Analysis runs only when you request it, is limited to the case you are in, and is recorded in the audit trail. The providers in use are listed in the sub-processor disclosure before any Customer Content is sent to them. We do not use Customer Content to train our own or anyone else’s shared models. Review: confirm provider terms on training, retention and location, and whether any provider processes data outside the UK/EEA
05Lana, the website help assistant
Lana on lexsalim.com answers questions about LexSalim using only published help-centre articles and product pages. It gives no legal, tax, accounting or financial advice, has no access to any account, case, document or payment, and cannot change anything. Where it cannot answer it says so and offers the contact page. Lana runs in your browser; conversations are not recorded on our servers. If live AI responses are enabled in future, this statement and the Privacy Policy will be updated first.
06Logging and evaluation
Agent requests, the documents they used and the resulting proposals are recorded in the case audit trail so that organisations can show who asked for what and when. We may use aggregated, non-identifying quality metrics (for example acceptance and rejection rates) to improve evaluations and policies. We do not read Customer Content for that purpose without the organisation’s agreement. Review: confirm whether any human quality review of outputs is permitted and under what consent
07Jurisdiction and scope limits
Legal research and tax features are released per jurisdiction after validation and are marked Planned or Early access until then. Where no jurisdiction pack is available, the AI works only from your documents and general knowledge and must not be treated as a statement of the law in any country. See the Legal AI Disclaimer and the Accountancy AI Disclaimer.
08AI regulation
We design the Service so that a person remains responsible for every decision. Where AI rules apply to the Service or to your use of it (for example the EU AI Act for customers in the European Union, or professional-conduct guidance on the use of generative AI for solicitors and accountants), you remain responsible for complying with the obligations that apply to you. Review: assess EU AI Act classification and transparency obligations, and relevant SRA, Law Society of Ontario and state-bar guidance on AI use
09Reporting problems
If output appears wrong or unsafe, reject it in the review queue and report it through the contact page. Reports help us improve evaluations and policies.
