Trust Center · Data & AI
Data handling and AI boundaries
Deterministic calculations stay separate from assistive AI. Exit Protocol may use extraction or AI tools to help prepare information for review. Those tools do not decide legal ownership, credibility, admissibility, or the deterministic LIBR arithmetic shown in an attorney-reviewable workpaper.
Last updated: 13 July 2026
| Stage | Possible processing | Required boundary |
|---|---|---|
| Upload | File validation, metadata, size and type checks, storage, and SHA-256 reference creation | The user must be authorized to process the record. |
| Text extraction | Native PDF text extraction or a local OCR runtime | Extraction may be incomplete and does not create an authoritative ledger. |
| External extraction | Document-processing provider only when explicitly enabled and when configured in your deployment | The customer or deployment administrator must approve the provider for the data class. |
| Candidate review | Suggested dates, descriptions, amounts, account hints, and confidence indicators | Human review before import, target-account confirmation, and reconciliation are required. |
| AI assistance | Drafting, summarization, classification, or review suggestions | AI output is editable assistance, not a factual or professional conclusion. |
| LIBR calculation | Deterministic ledger replay using selected records, source transaction, strategy, and assumptions | Generative AI does not calculate the traceable balance. |
| Workpaper export | Structured PDF, source references, calculation snapshot, and final file hash | A qualified reviewer must verify methodology, data, jurisdiction, and intended use. |
What Exit Protocol may store
- Account, case, party, role, and configuration data
- Uploaded evidence, extracted text, candidate rows, and review status
- Transactions, source links, assumptions, calculation state, and workpapers
- Hashes, access events, processing errors, integration identifiers, and support records
What AI does not decide
- Who legally owns property
- Whether a person is truthful, fraudulent, abusive, or credible
- Whether evidence is complete, authentic, privileged, or admissible
- Which tracing method or jurisdictional rule a professional should use
- Whether an output should be filed, disclosed, relied upon, or presented
Model-training policy
Exit Protocol does not use customer case content to train its own general-purpose models. Synthetic data, public information, or de-identified evaluation data may be used for testing and product quality.
When an external AI provider is enabled, its handling of submitted content is governed by the configured account, provider terms, retention controls, and regional settings.
User controls
- Use deployments with external AI or extraction disabled.
- Select the matter and document used in a connected workflow.
- Review, edit, reject, or approve candidate data.
- Disconnect integrations and revoke access.
- Request access, correction, export, or deletion subject to applicable limits.
Human review is a product requirement, not a disclaimer shortcut.
Reviewers should confirm the target account, statement coverage, source transaction, duplicates, opening balance, transfers, reconciliation, calculation strategy, jurisdiction, legal citations, and final file hash. A workpaper is a structured review artifact, not a substitute for professional responsibility.
This page describes product boundaries and is not legal advice, accounting advice, an expert opinion, or a security certification.