Security & responsible disclosure

The Criminal Defense Command Center serves people facing criminal charges, often during the most consequential period of their lives. A bug on our marketing landing is a bug. A bug that leaks matter data, bypasses authentication on a CDC route, or weakens the audit log can put a real person at risk of an outcome they cannot undo.

We will triage any disclosure within 72 hours. Findings that could undermine attorney-client privilege under the Kovel agency framework, or that could deanonymize an aggregate stat on the public Trust Center, are treated as critical and get same-day attention. We publish acknowledgments (with permission) and do not send legal threats to researchers acting in good faith.

These are the primitives that, if broken, would cause the most harm. Please read this list before you start.

• Kovel agency framework. Any vulnerability that could undermine attorney-client privilege on a Kovel-designated matter is treated as CRIT. This includes anything that exposes pre-engagement content as if it were privileged, or that leaks post-engagement content outside the attorney-agent umbrella.
• Matter data RLS. Postgres row-level security on the CDC matter tables. A bypass that lets one user read or write another user's matter row is CRIT.
• Audit-log immutability. The CDC audit log is append-only by RLS and SECURITY DEFINER. Any finding that lets a caller mutate, delete, or forge an audit row is CRIT.
• cd_attorney_packets storage bucket. Private bucket holding generated attorney packets. A finding that lets a non-owner read or list packet objects is CRIT.
• Trust Center k-anonymity floor. The public Trust Center at /trust/criminal-defense hides any aggregate count below the floor and returns "fewer_than_100" instead. If any number on that page can be deanonymized back to a single user or a single attorney, that is CRIT.
• Safe mode + quick exit. The safe-mode and quick-exit overlay is shared with Family Law, but CDC has its own surfaces that mount it (record clearing, bail, court date tracker). A finding that weakens the overlay on a CDC surface is treated the same as on Family Law.

We are most interested in findings that fall into one of these buckets:

• Bugs that could leak matter data to the wrong user, the wrong attorney, or out of the platform entirely
• Authentication or authorization bypasses on CDC routes, including record clearing, bail, court date tracker, and the attorney packet portal
• Anything that breaks the audit log: missing entries on a write path, mutability of existing rows, forgeable actor or matter UUIDs
• Cross-side exposures: attorney-side data leaking to consumers, or consumer-side data leaking to attorneys who should not see it
• Anything that weakens survivor-safety primitives (safe mode state, quick-exit overlay, shared-device detection) on a CDC surface
• /dashboard/criminal-defense/* consumer routes (dashboard, triage, intake, record clearing, bail, court dates, packet review)
• /api/cd/* consumer APIs (matters, triage, intake, eligibility screener, packet generation, attorney match)
• /attorney/dashboard/leads/criminal-defense/* attorney lead portal
• /admin/cd/* admin surfaces (unauthenticated probes only; do not access real admin data)
• The Trust Center at /trust/criminal-defense and its stats API

• Missing best-practice headers when there is no demonstrated impact
• Theoretical timing attacks without a working exploit
• Anything in node_modules or a third-party dependency we do not vendor
• Automated-scanner findings without a working exploit attached
• The marketing site at /criminal-defense and the per-state SEO pages at /criminal-defense/[state]. These read static or DB-backed content and do not perform writes; report a finding only if you can demonstrate an impact on the consumer dashboard or an API
• Denial-of-service via expensive LLM endpoints. These are rate-limited per user and per IP. A finding that bypasses the rate limiter is in scope; volume against the limited endpoint is not
• Social engineering targeting LawSensai employees, our attorneys, or our users
• Physical attacks against LawSensai facilities
• Findings whose only impact is on a researcher's own test account

We will not pursue legal action against researchers who:

1. Act in good faith and report directly to us
2. Do not access more user data than is needed to prove the issue, and do not modify, exfiltrate, or destroy other users' data
3. Do not degrade availability for other users
4. Give us a reasonable window to ship a fix before public disclosure
5. Comply with applicable law

Email security@lawsens.ai with:

• A clear description of the finding
• Steps to reproduce, ideally with a video or screenshots
• The affected URL or API endpoint
• Any proof-of-concept artifacts (no real user data)
• Whether you believe the finding affects the Kovel privilege posture, the Trust Center k-anonymity floor, or the audit-log immutability primitive, so we can route it on the critical lane

PGP key for sensitive reports is available on request. Tag privilege-relevant or survivor-safety-relevant findings with [PRIVILEGE] or [SAFETY] in the subject for same-day triage.

1. Acknowledgment within 72 hours (same day for [PRIVILEGE] or [SAFETY])
2. Triage decision + rough timeline within 5 business days
3. Fix shipped, then verification round-trip with you
4. Public acknowledgment (with your permission) once the fix is live

These are the same disclaimers we surface at point of capture inside the product. They are included here so any researcher reviewing user-facing copy can verify that the disclosures match the canonical source.

Triage surface: LawSensai provides legal information, document organization, and attorney matching. It is not a law firm. It does not replace advice from a criminal defense attorney.

Case Intelligence report: This report is an organizational summary. It is not legal advice, an opinion on the merits, or a prediction of outcome.

Pre-engagement privacy: This information is not protected by attorney-client privilege. Government investigators may be able to compel disclosure.

The Criminal Defense Command Center is scoped for an annual external penetration test by an independent third party. The test report's executive summary is published to this page once available; the detailed findings stay internal until remediated and re-tested.

When a researcher disclosure leads to a code change, we link the commit + PR here with the researcher's permission. The list below is intentionally short because the platform is young; we expect it to grow.

No public disclosures yet. Internal audit findings and their fixes are tracked in the repository underCriminalDefenseCommandCenter_QA_Audit.md and the V2.9 remediation sprint notes.