Trust & Readiness
Last updated: 2026-07-19
This page states what Matterline has actually built, what is not built yet, and what we will not claim. It is written for attorneys deciding whether this is safe to put in front of a matter. If anything here conflicts with what you see in the product, that is a bug — please tell us.
Current status
Internal development — fictional or redacted data only.
Matterline is not yet cleared for live, un-redacted client material. Several controls required for privileged data are not in place, and no third-party penetration test has been performed. We would rather tell you that plainly than let you find out later.
Where we are on the ladder
We assess our own readiness against five rungs. We are on the second.
Fictional data only
Internal development← we are here
Private beta with redacted data
Limited pilot
Ready for real client data
What that means for you
Available now
Look at it with us
We walk through a built-in fictional matter together. You tell us where the output is wrong, thin, or useless. None of your data is involved.
Available now
Try it on a redacted matter
Take a closed file, strip names and identifiers, and paste it. The product requires you to confirm the material is fictional, anonymized, or properly redacted before it will run.
Not yet
Live, un-redacted client matters
Blocked until the open items below are closed — signed data processing agreements, enforced MFA, key management, row-level security, and an independent penetration test. You will be told when that changes, in writing.
What is built and tested
Every item below is implemented and covered by automated tests, a live verification, or both.
Access & tenant isolation
- Sessions ride in HttpOnly, SameSite cookies (Secure in production) and are validated server-side on every protected request — never by trusting client state.
- A single authorization boundary. Identifiers and roles from the browser are never trusted; firm membership is re-fetched and compared on every request.
- Cross-firm access returns 404, not 403 — we never reveal that another firm's matter exists. Proven by tests, including URL-tampering attempts.
- Authentication is managed by Supabase. We do not implement our own password storage or cryptography.
How the AI step is handled
- All model calls run server-side only. The API key never reaches the browser.
- Prompts and model responses are never written to logs.
- Your notes are treated as source material, not instructions. A prompt-injection guard fences the content and neutralizes forged delimiters; suspicious input is flagged for extra review rather than silently trusted.
- Model output is schema-validated before it is shown or saved. Invalid output is rejected, not displayed.
- Pasted notes are not persisted by default — they are processed to build the timeline and discarded unless you choose to save the result.
Confidentiality of records & logs
- The audit trail records actions and identifiers only, against a metadata allow-list, with one-way hashed IP addresses. It cannot store case content.
- Logging redacts secrets and confidential fields recursively by field name.
- AES-256-GCM encryption for opt-in retained source notes, with tamper detection and context binding.
- Deletion workflows for analyses, matters, accounts, and firms, each with a confirming audit event.
Application hardening
- Security headers verified live: Content-Security-Policy, HSTS, nosniff, X-Frame-Options: DENY, referrer and permissions policies.
- Same-origin enforcement on state-changing requests (CSRF defense-in-depth), covered by tests.
- Per-user rate limiting on the AI endpoint, backed by memory or shared Redis.
- Continuous integration runs typecheck, lint, 56 automated tests, a production build, and a dependency audit on every change.
- Tenant isolation, unique constraints, encryption round-trip, and cascade deletes verified against a real PostgreSQL database.
What is not in place yet
These are the gates between today and live client data. None of them are hidden.
- Signed data processing agreements with our infrastructure and model providers, with confirmed data regions.
- Counsel review of the privacy policy, terms, AI disclosures, breach-notification process, and subprocessor list.
- Enforced multi-factor authentication. The provider and data model support enrollment; it is not yet mandatory.
- Managed key management and key rotation. Encryption is implemented; keys are not yet held in a managed KMS.
- Database row-level security as defense-in-depth behind the application's isolation checks.
- An independent penetration test, plus secret scanning in the build pipeline.
- A stricter Content-Security-Policy. The current policy still permits inline scripts and styles.
- Automated identity deletion tied to account closure, and litigation-hold handling.
- Scheduled retention enforcement and an external error-monitoring provider.
Who else touches the data
If you use Matterline, these providers process data on our behalf. Agreements are being put in place before any live client material is accepted. See Subprocessors for the full list.
TODO LEGAL REVIEW: Confirm the model provider's data-retention and training terms for our billing tier, in writing, before accepting any client material. Free and paid tiers differ materially on whether submitted content may be used to improve the provider's products.
Ask us directly about model-provider data-retention and training terms before you rely on them. We will show you the contract language rather than paraphrase it.
What we will not claim
Matterline makes none of the following claims, anywhere in the product or in our marketing:
If a vendor tells you otherwise about their own product, ask them for the report.
Your obligations, not ours
Two things remain yours regardless of what we build.
- Every timeline requires your review. Matterline produces a draft for an attorney to verify — it does not practice law, give legal advice, or produce filing-ready work.
- Your duty of confidentiality is unchanged. Your rules of professional conduct govern whether, and on what terms, you may route client information through a third-party vendor — including whether client consent is required. That judgment is yours, and we will give you whatever information you need to make it.
Questions
Email mateo@matterline.app. Fuller detail lives on our Security, Privacy, Terms, AI Usage, and Retention pages.