Authentication & access control
Workspace access is authenticated, role-aware, and scoped to the records a user is permitted to work with. Collaboration features are designed around least-privilege sharing rather than broad workspace exposure.
Trust Center / Security
Security-focused architecture for sensitive case operations.
Caseflow is built around authenticated workspace access, role-aware collaboration, document-scoped authorization, controlled AI processing, and reviewable operational workflows.
Operational security model
The security model combines application permissions, backend validation, scoped document access, provider-managed infrastructure controls, and human review requirements.
Data isolation
Case, task, document, and AI retrieval paths are designed around tenant and case boundaries. Shared access is rebuilt from current permissions before sensitive context is surfaced.
Document authorization
Supported document flows use storage paths plus permission-checked access requests. New migrated case and task uploads avoid storing reusable Firebase download-token URLs.
AI safeguards
AI requests go through backend controls, permission-aware retrieval, scoped vector filtering, stale artifact checks, confirmation flows, and human review requirements.
Auditability
Case history, AI audit metadata, confirmation records, and operational logs support traceability without treating logs as a substitute for professional review.
Monitoring & response
Support and operational channels route access, privacy, security, and AI handling concerns for review. Incident-response wording avoids unsupported SLA or certification claims.
Workspace isolation
Caseflow access checks are intended to keep each user inside the workspace, case, task, and document scope they are authorized to use.
- Tenant-scoped ownership and shared access records.
- Case-level capabilities for read, update, share, mailbox, task, and document workflows.
- Task assignment and task-grant paths for constrained collaboration.
- Accessible-document rebuilding before AI retrieval or controlled document access.
Encryption and storage
Data is stored with provider-managed infrastructure safeguards, while application controls govern who can reach records and files.
- Authenticated Firebase and backend routes for workspace operations.
- Storage-path based document records for migrated upload flows.
- Short-lived signed access URLs for supported path-backed document views.
- Legacy URL-only records retained for compatibility and tracked as residual risk.
Secure development and operations
- Security-sensitive operations are routed through backend validation where practical.
- AI write actions are allowlisted, risk assessed, and confirmation-gated.
- Document access improvements favor short-lived, permission-checked access over stored token URLs.
- Security information is kept practical and updated as the product changes.
- Known legacy URL and attachment surfaces remain tracked as migration work rather than hidden.
Shared responsibility
- Review workspace members and access grants regularly.
- Use role-aware sharing and document-scoped authorization for sensitive matters.
- Keep account credentials, devices, and browser sessions controlled.
- Validate generated documents, AI summaries, and workflow updates before external use.
- Report suspected unauthorized access, exposed links, or AI data handling concerns through support.
Responsible disclosure
Security reports, suspected unauthorized access, and privacy or AI data handling concerns should be submitted through support so they can be routed to the right reviewer. Dedicated vulnerability disclosure SLAs or bug bounty terms should only be published after operational approval.
Security issue
Include affected URLs, account email, observed behavior, screenshots where safe, and whether data may have been exposed.
Access concern
Report unexpected case, task, document, invitation, or workspace visibility with the minimum detail needed for triage.
AI handling concern
Share the case workflow, AI feature used, and why the generated output or retrieved context appears outside expectations.