Security
Security summary (PDF)How CaseRead keeps your documents private: where your documents live, who inside your firm can see them, who at CaseRead can reach them, and what happens to your data. Every control on this page is live in CaseRead today.
Compliance overview
Where CaseRead stands
Documents
For your file
Your documents, in a schema of their own.
Other firms’ searches can’t reach your vault.
Controls live today
Every control, stated plainly
Data isolation
- A database schema for every firm. Every firm gets its own database schema the moment it signs up, with its own document tables, search index and row-level security. Other firms’ searches can’t reach your vault, and vault search picks the schema from the signed-in lawyer’s own account, never from the request.
- A schema for each firm. Each firm’s document vault (file records and text, search index, projects, folders and version history) lives in its own database schema with its own row-level security, and the uploaded files sit in a private storage bucket, in a folder per firm.
- Not exposed through our API. Firm vault schemas aren’t reachable through our public database API. A two-firm attack test in August 2026 found no cross-firm access.
- What sits in shared tables. Conversations, chat attachments, project memory and calendar entries sit in shared tables protected by row-level security.
- Private by default. A private project is visible only to you and the colleagues you share it with, not to other firm members or firm admins. The database enforces it.
Access
- Two-factor sign-in. Every CaseRead account can turn on two-factor sign-in with an authenticator app, with recovery codes and a list of signed-in sessions you can end. A firm admin can require it for everyone in the firm.
- A firm activity log. Firm admins see who viewed, downloaded or exported a document, who was denied one, and every role change, invite and two-factor change, and can export the log to CSV. A private document’s title stays hidden from anyone who can’t open it.
Encryption and hosting
- AES-256 encryption at rest. Our database host, Supabase, encrypts everything it stores for us at rest with AES-256.1 CaseRead encrypts Google and Microsoft connection tokens and share-link passwords again with AES-256-GCM.
- Encrypted in transit. Every connection runs over HTTPS (TLS), and every page on www.caseread.ai sends a two-year HSTS header. Keys you issue to connect Claude, ChatGPT or Word are stored only as one-way hashes.
- Hosted in the US. Your firm’s database and files are hosted by Supabase on AWS in the United States.
Your data
- No training on your data. CaseRead runs no model training or fine-tuning on your data, and our AI providers’ commercial terms exclude it from their training.
- Instant, verifiable purge. Delete a project and CaseRead erases its documents, files, versions, AI conversations, memory, workflow results, monitors, share links and upload links, across every CaseRead store and for every member of the firm, in one step. We keep a content-free record that the purge ran: counts, never names. Events already sent to your own calendar stay there, and our AI providers’ own retention terms still apply.
- Closing an account. When an account is closed or disabled, its sessions and connected-app keys are revoked. If you’re a solo firm, closing your account deletes your firm’s document vault and chats (file removal is best-effort). In a larger firm, it removes your access and the firm keeps its documents.
Citations
- Every citation checked. The Hallucination Shield runs two checks on every citation you give it: does the source exist, and does it support the claim it’s cited for.
Sources
- Supabase, “Security”: “All customer data is encrypted at rest with AES-256 and in transit via TLS.” Read September 14, 2026. supabase.com/security
- ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512 (July 29, 2024). Read September 28, 2026. americanbar.org
The companies that process data for us are listed in our Privacy Policy.
Inside your firm
Who in your firm can see what
Private stays private, even from firm admins.
- Private by default. New projects are private to the lawyer who creates them. A private project is visible only to you and the colleagues you share it with, not to other firm members or firm admins. The database enforces it.
- Share on purpose. Keep a project private or share it with your firm.
- Three roles. Admin, member and viewer. Viewers are read-only: they can’t upload, edit, share or run research. Only admins can remove members.
- Two-factor sign-in. Every CaseRead account can turn on two-factor sign-in with an authenticator app, with recovery codes and a list of signed-in sessions you can end. A firm admin can require it for everyone in the firm.
- A firm activity log. Firm admins see who viewed, downloaded or exported a document, who was denied one, and every role change, invite and two-factor change, and can export the log to CSV. A private document’s title stays hidden from anyone who can’t open it.
Inside CaseRead
Who at CaseRead can reach your data
Our engineers hold database credentials that could reach stored data. These are the controls on our admin console; direct database access sits outside them.
- No document screens. Our admin console has no screen that shows your documents or chats.
- Impersonation is logged before it starts. Support impersonation is written to an append-only audit log first. If the log can’t be written, it doesn’t start. Impersonation sessions expire after 15 minutes, and staff accounts can’t be impersonated.
- High-risk actions, logged first. The same audit-first rule covers firm and user deletion, data purges, password resets, invite links and bulk email.
- MFA on our admin console. Our staff admin console requires a TOTP code every session. Impersonating an account, resetting a password or deleting a firm requires one entered in the last 30 minutes.
- Usage records carry no document names. The usage and cost records our admin console reads never contain your document names or the text of your questions and answers.
AI and your data
Your data isn’t used to train AI models.
CaseRead runs no model training or fine-tuning on your data, and our AI providers’ commercial terms exclude it from their training.
- CaseRead sends each task to the services that do it: Anthropic for reasoning, drafting, OCR and web search; OpenAI for the search index; and the public-law sources, which receive search queries. AI providers receive your data only for tasks you or your firm start, including workflows you schedule.Your questionCaseReadClaude, embeddings, public law
Ethics
Rule 1.6 and ABA Formal Opinion 512
ABA Formal Opinion 512 (July 29, 2024) says a lawyer needs the client’s informed consent before entering client information into a self-learning AI tool, and that boilerplate in an engagement letter is not enough.2
Your Rule 1.6 confidentiality duty, mapped to CaseRead’s controls.
- Keep your firm’s client files away from other firms. Each firm’s documents live in their own database schema, with row-level security inside it.
- Limit who inside the firm sees a matter. New projects are private to the lawyer who creates them.
- Stop a stolen password from opening the account. Two-factor authentication for every account, and a firm admin can require it firm-wide.
- Know who opened, downloaded or exported what. A firm activity log of document access, role changes and invites, with CSV export.
- Keep client information out of AI model training. Your data never trains AI models: CaseRead trains none, and our AI providers’ commercial terms exclude it.
- Erase a client’s files when the matter ends. Purge a client’s project instantly across every CaseRead store, with a content-free record that it ran.