Skip to content
LQ.AI

Data, privacy and evidence

  • evaluator
  • operator
  • partner

Understand where your information goes, who can access it, and what the app records. You are here because someone asked you to sign off on LQ.AI, or to hand it to someone who will. This page is the index of every artifact that answer will need — what it covers, what it deliberately leaves to you, and the file in the repository it is checked against. Every table row links to a page that curates the underlying artifact rather than restating it, so what you read here and what the code does stay the same document.

LQ.AI is self-hosted: there is no LegalQuants-operated service standing between your deployment and this evidence. Every claim below traces to a file you can open in the same repository that built the running system — see Verify these claims yourself for the commands.

The artifacts

Section titled “The artifacts”

What information is sent outside LQ.AI?

What it covers
Where chat and skill content goes, tier by tier, and what a miss looks like
What stays the operator’s
Which inference tier you route each matter to
Canonical file
docs/security/anonymization.md

Review the security risks

What it covers
STRIDE-by-component threats and mitigations for the five production services
What stays the operator’s
Your host, OS, IdP, and secret custody
Canonical file
docs/security/threat-model.md

Replace identifying details before sending a request

What it covers
What the pseudonymization layer catches, skips, and has (and has not) been measured against
What stays the operator’s
Recognizer tuning for your own document conventions
Canonical file
docs/security/anonymization.md

What the audit log records

What it covers
What is logged, what is not, retention, and the one-query evidence pattern
What stays the operator’s
Tamper-evidence infrastructure, retention policy
Canonical file
docs/security/audit-logging.md

Check the software you download

What it covers
Signed images, SBOM, dependency-update cadence
What stays the operator’s
Scanning the SBOM and applying patches
Canonical file
docs/security/releases/README.md

Known limitations

What it covers
Shipped / partial / scaffold / deferred, with a verification path for every row
What stays the operator’s
None — this is the honest inventory
Canonical file
docs/HONEST-STATE.md

Report a security problem

What it covers
How to report a vulnerability, and how the project decides public vs. private handling
What stays the operator’s
None
Canonical file
SECURITY.md

Answer security and procurement questions

What it covers
A pre-filled SIG Lite starter for privileged-matter handling
What stays the operator’s
The rest of your procurement process
Canonical file
docs/procurement/sig-lite.md

Match requirements to controls and evidence

What it covers
Framework-alignment status (SOC 2, ISO, GDPR, HIPAA, FedRAMP) and what’s still planned
What stays the operator’s
Your own certification
Canonical file
docs/compliance/README.md

Release history

What it covers
What each released version changed, and the upgrade class it carries
What stays the operator’s
Deciding when to re-review
Canonical file
docs/releases/

Check the claims yourself

What it covers
Commands and files to check every claim on this page yourself
What stays the operator’s
—
Canonical file
README.md

Each canonical file is named as a repository path; every page’s footer stamp links it at the commit that page was checked against.

The URL-stability commitment

Section titled “The URL-stability commitment”

Once a page under /trust/ ships, its address does not change. You can cite a specific URL in a security questionnaire, an insurer’s file, or a signed-off procurement record, and it will still resolve a year later — the page’s content may be revised (the footer stamp on every page names the commit it was last checked against), but the location does not move. If a URL you cited from this tree stops resolving, that is a bug against this site, not an expected consequence of a release.

Next

Section titled “Next”