What’s live, what’s gated, what still needs setup.

Security posture

Auth redirects run at the framework edge for dashboard, admin, and login paths. Route handlers own the detailed controls: CSRF/origin checks, rate limiting, webhook signatures, tenant checks, cron secrets, portal sessions, and admin role gates.

Security response headers are configured at the Next.js layer. Provider error logging is designed to avoid full prompt or customer-context echo.

• Route protection inventory is maintained as a build-time artifact.
• Public API routes must be intentionally classified.
• Card details are entered on Stripe's hosted checkout — Sovereign servers never store card numbers.
• No third-party certification is claimed unless the audit is complete.

Data retention

Operational records such as leads, conversations, audit logs, payments, consent evidence, service evidence, and AI usage records are retained while needed to operate the account, support billing, resolve disputes, debug incidents, and preserve proof history.

Deletion requests move through the account privacy flow. Some records may remain for legal, security, fraud-prevention, tax, billing, or backup-retention reasons.

Privacy and data requests

Tenant data is not sold. Customer business data, founder platform metrics, and AI prompt context are treated as separate scopes.

Account export and delete-request flows exist for authenticated customers. Privacy requests can also be sent to privacy@trysovereignai.com.

Account deletion runs in two phases: a deletion request marks the account, and after a 30-day grace period an automated job permanently deletes the account record, writing an audit-log entry that documents the hard delete itself.

AI disclosure

Customer-facing voice, chat, and automated messages are AI-assisted when AI is involved. High-impact outputs should identify that assistance in the surrounding customer experience.

AI-answered phone calls open with a spoken disclosure — “Just so you know, you're speaking with an AI assistant.” — on both inbound and outbound calls. The disclosure wording lives in one shared module so every voice path uses the same reviewed copy.

Business profile fields, CRM records, transcripts, and imported content are treated as untrusted context for AI prompts. They may inform an answer, but they do not become instructions to the model.

• Customer data-backed answers should distinguish facts from assumptions.
• Owner approval is required before high-risk outbound, reputation, money, or legal/trust actions.
• Autonomy increases only after quality evidence and customer approval.

Human approval and autonomy

Agent actions pass through policy and entitlement checks before side effects. Low-risk internal work can run on its own. Explicitly configured chatbot replies, reminders, review requests, and lifecycle messages may also run automatically; broad outbound, destructive, legal/trust, and money-moving actions use stricter approval or hard-block policies.

Approval-backed actions create an account-scoped request that records the proposed action, payload, reviewer, and decision time. The activity trail and workflow configuration are the authoritative record of whether a particular action was automatic or approved.

Consent and communications

Sovereign records consent evidence where forms, chatbot, API, discovery conversion, or customer workflows collect email, phone, or outreach permission.

Marketing email and SMS workflows must support unsubscribe and opt-out handling. Suppression lists, quiet-hour checks, and owner approval gates are part of the operating model for outbound communication.

This page is not legal advice. Customers remain responsible for their own regulated outreach, privacy notices, and industry-specific obligations.

Call recording

Voice workflows should use approved call recording disclosure copy before recording is enabled. Disclosure requirements can vary by location and call type.

Recording is off by default and turns on only when the business has explicitly granted recording consent. When recording is on, callers also hear “This call may be recorded for quality.” — the platform never says a call may be recorded when it is not, and never records without saying so.

Call recordings, summaries, transcripts, and AI coaching outputs are treated as customer account data and should be handled with the same tenant-boundary controls as CRM records.

Proof and claims

Sovereign does not invent testimonials, customer names, revenue, bookings, reviews, or integration status. Case-study metrics require source artifacts and permission.

Demo, beta, setup-required, and planned capabilities stay labeled until customer data and production evidence support stronger claims.