AI Disclosure

1. Where AI is used

• Transcription and diarization — converting audio and video to text with timestamps and separated speakers.
• Translation — cross-language search and exports between English and Spanish.
• Document and visual evidence extraction — reading document passages and visibly supported facts in screenshots, scans, photographs, and embedded exhibits while keeping them linked to source and page.
• Cited answers (RAG) — retrieving relevant excerpts and drafting answers that cite the exact recording, speaker, and timestamp, and that say so when the record does not support an answer.
• Inconsistency analysis — comparing qualified assertions and visible facts across recordings and documents, including differences in dates, amounts, attribution, and whether an action was proposed, recommended, or completed. Every flag retains both sources and context.
• Chronologies — extracting dated events and ordering them, traceable to source.

2. Known limitations

Automated outputs may contain errors in words, speaker labels, dates, numbers, names, translations, and inferences, especially with poor audio, overlapping speech, accents, jargon, or ambiguous context. Inconsistency analysis is deliberately conservative: it is a prompt for a human reviewer, not a finding of fact. Absence of a flagged inconsistency does not mean statements agree, and a flag does not mean a statement is false. OCR and visual interpretation can also misread compressed screenshots, handwriting, cropped context, or ambiguous interface elements.

3. Human review and boundaries

You must verify every material statement against the cited recording or document before relying on it. Tracecanon does not provide legal advice, does not authenticate recordings, does not determine admissibility, does not produce certified transcripts, and does not determine the truthfulness of any statement. AI is an aid to professional judgement, not a substitute for it.

4. No training on your content

We do not use Customer Content to train our own models. Protected-cloud requests go only through OpenRouter with data collection and training denied, explicit zero-data-retention routing, and an approved downstream-provider allow-list. See the Privacy Policy and Sub-processors list for detail.

5. Provider routing

Transcription and analysis use protected OpenRouter routing, restricted to approved providers with model training and data collection denied. A self-hosted Private mode is not currently offered; every matter uses this protected cloud routing. Machine translations carry a label reminding you to verify against the original-language source: “Machine translation — verify against the original-language source.”.

6. The notices you will see

These notices appear persistently in the app and on exports: