Measures common to the whole TODO.LAW stack →
Security & Trust
AI SENTINEL is built for organizations that manage sensitive AI governance data. Security is embedded in every layer of the platform: from how we isolate your data to how we control access and protect your information.
Data Isolation
Every organization's records are kept apart from every other organization's. Before the application shows or changes your AI systems, assessments, policies, incidents or vendor data, it checks that you are a member of your organization.
Organization-Scoped Queries
Queries for your governance records are filtered by your organization in each request handler.
Verified Membership
Before any operation on your organization's data, we verify that you are an active member of that organization. Membership is checked on every such request.
Cross-Tenant Protection
When a record points to another, the server checks that the other record is your organization's: a policy, data flow, oversight gate, incident, assessment, sensitive-data analysis or shadow AI report linked to an AI system; an AI system or data recipient linked to a vendor; and an assessment's template, which must be a shared template or one of yours.
Shared Reference Data Only
Only reference data (compliance frameworks, assessment templates, and the vendor and AI tool catalogs) is shared across organizations. Your governance data is never shared.
Access Control
AI SENTINEL enforces role-based access control (RBAC) on the server. Each change to governance records is checked against your role before it is carried out.
Read-Only Enforcement
Viewers have read-only access to governance records. Writes to those records are blocked at the server level, not just in the interface, so they cannot be made through API calls either.
Elevated Permissions
Approving assessments, approving or publishing a policy, and passing, failing or deferring an oversight gate require the AI Officer, Admin or Owner role, whichever screen or API call makes the change. Billing actions are limited to Owners and Admins.
Authentication
We support standard sign-in methods, with 12-hour sessions and rate-limited sign-in on the hosted service.
Google OAuth
Sign in with a Google account (OAuth 2.0)
Magic Link
Passwordless sign-in through a single-use link sent by email
Cross-Platform SSO
Single sign-on across the TODO.LAW apps (hosted service)
Session Security
- •HTTP-only cookies prevent client-side access to session tokens
- •Secure cookie flag whenever the app is served over HTTPS
- •Sign-in requests carry a CSRF token
- •SameSite=Lax session cookies limit cross-site request forgery
Input Validation
All user inputs are validated on the server before reaching business logic. This protects against injection attacks and ensures data integrity.
Schema Validation
Every application (tRPC) input is validated against a strict schema, and integration endpoints check their fields in code. Invalid data is rejected before it reaches the database.
Parameterized Queries
All database queries use parameterized inputs, including the few written in SQL, which bind every value. This removes the risk of SQL injection.
Enum Enforcement
Status fields, risk levels, and categories use strict enum validation. Only predefined values are accepted.
Audit Trail
Changes to governance records and key decisions are logged to an audit trail. This supports compliance requirements and gives a record of governance activities.
What We Log
- • AI system lifecycle changes
- • Risk classification decisions
- • Assessment submissions and approvals
- • Oversight decisions
- • Policy changes and publications
- • Incident creation and status changes
- • Organization membership changes
Retention
Audit records are kept when a member leaves the organization. When an organization is deleted, its records are deleted with it, including its audit history; one entry recording the deletion is kept.
Transport & Infrastructure
Your data is encrypted in transit with TLS, and at rest by the hosting and database providers' storage encryption.
HTTPS Everywhere
All connections are encrypted with TLS. HSTS headers ensure browsers always use secure connections.
Clickjacking Protection
Security headers prevent the application from being embedded in frames on other sites.
Content Security
MIME-type sniffing prevention and strict referrer policies protect against content-based attacks.
Minimal Permissions
Browser APIs (camera, microphone, geolocation) are explicitly disabled: the application never requests unnecessary access.
Where the data is stored
On the TODO.LAW cloud, if an admin turns AI features on, the text they use goes to the AI provider set up on the service. On a managed instance or on your own hardware, the data stays there, unless you connect an external AI provider.
API Security
Every endpoint that reads or changes organization data requires authentication. External integrations use signature verification and key-based authentication.
Endpoint Protection
Every governance endpoint used by the application requires an authenticated session with verified organization membership. Unauthenticated requests are rejected.
Webhook Verification
Inbound integration requests are authenticated with an API key, and webhooks use cryptographic signature verification. Unverified requests are discarded before processing.
Security Standards
AI SENTINEL's security controls are designed to address the OWASP Top 10 web application security risks.
Security Contact
If you discover a security vulnerability or have questions about our security practices, please contact us at security@todo.law. We take all reports seriously and will respond promptly.