exhibit.law
AI Disclosure
Last updated 2026-08-06
exhibit.law uses AI to make evidence usable faster: transcribing audio/video, running OCR on documents and images, identifying speakers, and drafting event summaries for your review. This page explains what runs where, and what that means for your case data.
You acknowledge this disclosure when you create an account. The contractual rules for AI outputs and human review are also in the Terms of Service.
How AI is used on exhibit.law
Uploaded media may be processed automatically to extract text, transcribe audio and video, identify speakers, run OCR or vision on documents or images, draft suggested timeline events, and expand zip archives. Those outputs are assistive and remain marked for human review until someone on your team confirms them.
Primary configuration: closed, self-hosted models
By default, and under normal load, AI processing on exhibit.law runs on closed, self-hosted models operating on infrastructure we control. Under that primary path, your case content for those features is not sent to third-party AI providers, and is not used to train any model, ours or theirs.
Peak-demand exception: third-party frontier models
During periods of peak demand, certain features (transcription, OCR, or drafting) may be processed using third-party frontier AI models instead of our self-hosted infrastructure, in order to maintain reasonable turnaround times.
When this occurs: the specific content sent is limited to what is needed to perform the requested function (for example, a transcript request sends the audio or video file, not your whole case); we use API / enterprise-tier arrangements under which your content is not used to train the provider’s models and is retained only as needed to process the request; routing is automatic based on system load — you are not asked to approve it feature-by-feature, but you acknowledge this possibility when you sign up.
Enterprise customers can request a dedicated configuration that disables frontier-model fallback entirely, keeping all processing on self-hosted infrastructure regardless of load — contact us via the contact form for details.
Current subprocessors for AI processing
Operators that may process case content for AI features:
- Self-hosted / primary — transcription, OCR/vision, drafting under normal load. Not retained beyond processing; not used for training.
- OpenAI (API / enterprise tier) — same features during peak-demand fallback only. Retention and training per OpenAI’s API data-processing terms for that tier (not used to train models under that relationship). Only content needed for the job is sent.
Why this matters for your case data
Case content on exhibit.law is frequently privileged, confidential, or sensitive (client communications, medical information, information about minors, criminal case materials). Sending that content to any third party — even briefly, even under a no-training agreement — is a disclosure that could be legally relevant to privilege or confidentiality obligations your organization owes to your own clients. That is why we disclose this here rather than folding it silently into general “AI processing” language, and why we ask you to acknowledge it explicitly at signup.
Human review, always
Regardless of which model processes your content, all AI output — transcripts, OCR text, speaker labels, drafted event summaries — is assistive only. It is marked as unreviewed until a person on your team confirms it, and should never be treated as a verified statement of fact about the evidence. See the Terms of Service, “AI outputs and human review,” for the contractual version of this.
Contact
Questions about AI processing: use the contact form at https://exhibit.law/contact.