AI Privacy and Data Handling

Caseflow uses AI to assist with search, extraction, document intelligence, summaries, drafting, transcription, speech, workflow guidance, and proposed actions. AI is not an autonomous professional decision-maker and does not replace a lawyer, investigator, fiduciary, accountant, medical professional, or other qualified reviewer.

• A signed-in user invokes an identified AI feature.
• Backend services verify the user, tenant, case, task, and document permissions relevant to the request.
• Caseflow prepares limited context, applies configured redaction and size controls, and sends it to OpenAI through the API.
• The response is returned as assistive content. Supported write actions require an allowlisted operation and, for higher-risk actions, a short-lived confirmation bound to the user and context.

Depending on the feature, inputs may include prompts, case and task fields, document text or files, images, contact or timeline information, extracted facts, and audio. Operational metadata may include hashed tenant or case identifiers, user and feature identifiers, model, token counts, redaction status, timestamps, safety status, purpose, and retention expiry.

Caseflow currently uses the OpenAI API. OpenAI states that API data is not used to train or improve its models by default unless the customer explicitly opts in. Standard abuse-monitoring logs may retain customer content for up to 30 days unless longer retention is legally or safety-required. Responses API application state may also be retained when the request or account configuration stores it. Zero Data Retention and regional processing are special configurations and are not promised unless expressly confirmed for the deployment.

• Accessible document sets are rebuilt from current permissions before hardened retrieval.
• Tenant and case metadata is scoped, with hashed identifiers where provider metadata is used.
• Retrieved vector results are post-validated and stale or superseded artifacts are rejected on the hardened path.
• Context budgets, upload limits, content validation, and production raw-upload safeguards reduce unnecessary exposure.
• Pattern redaction reduces common personal-data exposure but is not guaranteed anonymization.

• The interface identifies AI-assisted features and generated output.
• Users must check source records, citations, extracted facts, deadlines, calculations, and proposed actions.
• Workspace administrators should train users on capabilities, limitations, confidentiality, prompt handling, bias, and escalation appropriate to their role and risk.
• Do not use Caseflow AI for prohibited practices or solely automated decisions with legal or similarly significant effects.

Caseflow AI audit metadata has a 30-day default engineering retention setting. Case deletion is designed to remove related chat sessions, audit records, temporary artifacts, and index references. Saved outputs follow the record where they are saved. Provider-side logs and application state follow the applicable OpenAI configuration and contract.

AI can produce incomplete, incorrect, biased, or outdated content. It may miss context or misread documents. Report unexpected data access, unsafe output, or a suspected permission issue through support with the minimum sensitive detail needed for investigation.