Live implemented and verified in the client portal   Ops-managed enforced in the operations platform   Rolling out in active development

Domain 1

Identity and Data Protection

Participant identity vault

Live

Identifying fields are split off at collection into a table readable only by the client organization's own logins, enforced by database row level security. Operators, AI systems, and support staff have no read path. Verified by query.

Coded working records

Live

Every downstream record carries only a code (for example, P-4X7K2). Deliverables never name a participant.

Application layer vault encryption

Live

Identifiers (name, email, phone) are AES-256-GCM encrypted with a server-held key before storage, and decrypted only server side for the client's own view.

Server side intake screening

Live

Briefs, notes, uploads, and self serve AI inputs are screened for participant identifiers. A hit is rejected before storage: nothing lands, even briefly.

Unscreenable file honesty

Live

PDFs and images cannot be content screened, so uploading one requires an explicit confirmation that it contains no participant identifiers.

Domain 2

Authentication and Sessions

Passkeys (WebAuthn)

Live

Passwordless, phishing resistant login with fingerprint, Face ID, or device PIN.

Two factor authentication (TOTP)

Live

Opt in authenticator app second factor with a server side step up guard. Mandatory for operator accounts.

MFA recovery codes

Live

Stored hashed.

Breached password prevention

Live

A length and variety floor plus a HaveIBeenPwned k-anonymity check. The password never leaves the device.

Idle timeout sign out

Live

Automatic sign out after 30 minutes of inactivity.

Global session revocation

Live

Sign out of all devices, in one action.

Domain 3

Access Control and Isolation

Row level security everywhere

Live

Every table of client data is organization scoped at the database.

Role based access

Live

Client admin, read only viewer, and operator roles.

Assignment scoped operator access

Ops-managed

Operator access narrows to assigned organizations, with flagged, logged, non silent administrative overrides.

Service role isolation

Live

The one role database rules do not bind is restricted to our own server code.

Cross client AI isolation

Live

A client's AI context is drawn only from that client's own brief and brand.

Domain 4

Application Security

Content Security Policy

Live

Enforced CSP with a per request nonce and strict-dynamic, blocking injected and inline scripts, with violation reporting.

HTTP security headers

Live

HSTS (two year, preload), X-Frame-Options DENY, X-Content-Type-Options, Referrer-Policy, and Permissions-Policy.

Rate limiting

Live

Durable, fail open rate limits on sensitive actions.

Private file storage

Live

Non public buckets with short lived signed URLs only.

CSRF protection

Live

Framework server actions with origin verification.

Coordinated disclosure

Live

A published security.txt (RFC 9116).

Domain 5

AI Governance

No training on client data

Live

Committed under our AI provider's commercial API terms.

Human delivery gate

Live

AI has no technical capability to deliver work to a client or any third party.

No AI path to the vault

Live

AI systems cannot read participant identities.

Honest output standard

Live

Measured results labeled measured, judgment labeled judgment, gaps flagged rather than invented, participants by code.

Domain 6

Audit, Logging, and Monitoring

Tamper evident audit log

Live

Each record is sealed in a SHA-256 hash chain, and a verifier reports the first break.

Activity timeline

Live

Clients see the actions taken on their requests (submitted, produced, reviewed, delivered, approved). The timeline states plainly that it shows actions, not reads.

Recorded data access log

Rolling out

A metadata record of each read (actor class, data class, timestamp, never content), surfaced to the client.

Confidentiality before access

Live

Written confidentiality obligations are signed before any access is granted.

Domain 7

Infrastructure and Data Lifecycle

Encryption in transit and at rest

Live

TLS in transit; AES-256 at rest, including backups.

Patched, current stack

Live

A current, patched framework with active dependency vulnerability management.

Free data export

Live

Any time during the term, and for 30 days after.

Deletion on exit

Live

Active systems within 60 days of termination, backups within 90 days, confirmed in writing.

Breach notification

Live

A 72 hour notification commitment, per the Data Processing Agreement.

Retention schedule

Live

A defined data retention schedule with legal holds.

Domain 8

Governance and Documentation

Documents match the platform

Live

The Privacy Policy, Terms, DPA, AUP, AI Disclosure, and Subprocessor Disclosure are kept consistent with what the platform actually does.

United States personnel only

Live

All access to client data is by our own personnel, in the United States, under written confidentiality obligations. No offshore talent works on the platform or accesses client data.

Native electronic signature

Live

Agreements are signed in platform and sealed with a content hash and a certificate of completion.

Accurate posture

Live

We describe controls as implemented and verified, and we do not claim certifications we do not hold.

Subprocessors

All United States, All Disclosed

ProviderFunctionLocation
SupabaseDatabase, authentication, file storageUnited States
VercelApplication hostingUnited States
AnthropicAI draftingUnited States
StripePaymentsUnited States
ResendTransactional emailUnited States
TwilioSMSUnited States
NetlifyMarketing websiteUnited States

The full list, with what each provider can reach and our 30 day change notice, is in the public Subprocessor Disclosure.

Honest limits and roadmap

What we claim, and what we do not

  1. Access separation, not client held cryptography. Database rules and a server-held key make the vault unreadable by our roles, and a stolen backup yields only ciphertext. But our infrastructure includes a server side role and a server-held key restricted to our own code. Our claim is that our people and our systems cannot read participant identities, verified by query. We do not claim that no party could mathematically do so.
  2. Screening is a safeguard, not a guarantee. Intake screening is pattern based, and it is backed by the Acceptable Use Policy rather than replaced by it.
  3. Recorded reads are rolling out. Today the client view shows the actions taken on requests, and says so plainly.
  4. Roadmap, stated as roadmap. SOC 2 Type II and an independent third party penetration test are on our near term roadmap. Neither is complete, and we will not describe either as achieved until it is.
The Paper

The Documents Behind This Page

Privacy Policy · Terms of Service · Data Processing Agreement · Acceptable Use Policy · AI Disclosure and Consent · Subprocessor Disclosure

Full architecture detail, including the participant identity vault and how we use AI, is on the Trust Center. Security questions and vulnerability reports: support@lilxhub.com with Security in the subject line.