AI Transparency & Disclosures

You are interacting with an automated system on the surfaces below. There is no human reviewing output before you see it.

The current provider list, including what each one processes and where it operates, is published on our Subprocessors page and is subject to the change-notice commitment in our DPA.

We do not train models on customer data, and we do not permit our model providers to. Prompts, documents, conversations, and generated output are sent to providers under enterprise terms that prohibit training on the content and prohibit sharing it with other customers.

Providers may retain content transiently for abuse monitoring and operational purposes under those terms. We do not sell customer data and we do not disclose it for advertising — see Your Privacy Choices.

These are properties of the technology, not defects we expect to eliminate:

• Fabrication. Models can invent case names, citations, quotations, statutory sections, and holdings that do not exist, and present them in correct-looking form.
• Staleness.A model's general knowledge ends at its training cutoff. Retrieved sources may themselves be superseded, appealed, repealed, or amended.
• Jurisdictional drift. Absent explicit instruction, a model may apply the law of the wrong jurisdiction, or blend jurisdictions, while sounding certain.
• Incomplete retrieval. Semantic search returns what is similar, not what is exhaustive. Absence of a result is not evidence of absence of authority.
• Context limits. Very long documents may be processed in parts; a conclusion drawn from one part may miss a controlling provision in another.
• Non-determinism. The same prompt can produce different output on different runs. Reproducibility is not guaranteed.
• Bias. Models reflect patterns in their training data, including patterns that are skewed or discriminatory.

Every output must be reviewed by a qualified person before it is relied on, delivered, or filed. Citations must be checked against the primary source. Figures, dates, defined terms, and party names must be checked against the underlying document. Your obligations of competence and supervision are described in the Legal Disclaimer.

We do not use AI to make decisions that produce legal or similarly significant effects about individuals — we do not score, profile, rank, or screen people, and we do not use AI to decide who may open an account or what they are charged. The models act on the material you submit, at your instruction, and return text to you.

If you use the Service to inform decisions about individuals — hiring, credit, housing, insurance, benefits, or similar — you are the decision maker and you are responsible for any assessment, notice, appeal, or human-review obligation that applies to that decision under the law governing you.

Material produced by the Service is machine-generated. Where the law governing you, a court, an agency, or a client requires that AI-assisted work be disclosed or labelled, that obligation is yours; the Service does not watermark exported files or apply disclosure language on your behalf.

As between you and us, you own the output generated from your material, on the terms set out in the Terms of Service. Note that generated text may not be independently copyrightable, and that similar prompts may produce similar output for other customers. We make no representation that output is free of third-party rights.

We change models, versions, and providers as the technology moves. Behaviour, phrasing, latency, and quality may change without notice for changes that do not affect who processes your data. Changes to which subprocessor processes customer data follow the notice procedure in the DPA.

Tell us when the system gets something badly wrong — particularly fabricated citations, unsafe content, or output that appears to reproduce someone else's protected material. Email support@elawplatform.com, or abuse@elawplatform.com for misuse of the platform by another user.