Trust, security and responsible AI

Every organization is a tenant. Tenant data is separated at the database layer with row level security, so a request running for one tenant cannot read another tenant’s rows. Isolation is enforced where the data lives rather than only in application code. Within a tenant, users, roles, permissions and workspaces limit what each person can see and change.

Isolation is built and tested. We recommend an independent penetration test before regulated customers rely on it, and we will say plainly when one has been completed.

• Each tenant has its own data encryption key, wrapped by a master key. Call audio is sealed at rest with AES-256-GCM under the tenant key.
• When a tenant is erased, its key is destroyed, which makes sealed content unreadable. This is how erasure requests are honored.
• Authenticator-app multi-factor sign-in with single-use recovery codes. Secrets are stored encrypted and recovery codes as hashes.
• Enterprise sign-in through OIDC and SAML, with SCIM provisioning for identity providers.

Encryption in transit depends on how the service is deployed. We do not publish a transport-security claim on this page beyond what the platform itself enforces.

Significant actions are written to a tamper-evident audit trail, so records of who did what cannot be quietly altered. Assistants act within action-level permissions, and high-impact actions can require human approval through a named approver, with the decision recorded.

Data sovereignty matters to many organizations, and we treat it as a product requirement. Regional data residency and a supported self-hosted deployment are on our build roadmap and are discussed during the engagement for customers who need them. Tell us your jurisdiction and constraints through the contact form and we will answer specifically about what applies to you.

Assistants identify themselves as AI. A person speaking with a WoneAssistant assistant is never led to believe they are speaking with a human. The platform includes a disclosure rule that can be configured per jurisdiction, including whether the disclosure is spoken at the start of the conversation, and assistants hand off to a person when asked.

Outbound calling uses legitimately assigned or verified caller identification, never spoofed numbers.

General-purpose voice calling through WoneAssistant must not be treated as emergency-call capable unless the applicable service requirements have been met for your country, your numbers and your endpoints. Emergency calling needs a validated address for each number, a supported endpoint and a carrier that supports it.

The platform includes emergency calling handling: addresses are validated and bound to each number, and emergency numbers are matched ahead of every other check and connected straight to the carrier. This has been tested in a laboratory setting and is not yet verified on live telephone networks. Until we confirm it for your setup in writing, keep a separate way to reach emergency services. Never rely on an assistant to place or relay an emergency request.

We make only the statements we can back with the product as it exists.

• A compliance engine applies consent, calling hours, AI disclosure and retention rules to calls and messages. Legal review is required per jurisdiction.
• GDPR export and erasure rights are supported for personal data.
• We do not currently claim SOC 2, ISO 27001, HIPAA or PCI certification or attestation. If you need one, tell us and we will say where we stand.

The providers that process data for us are listed in section 4 of the Privacy Policy: Cloudflare for delivery and DDoS protection, Stripe for payments, hosting infrastructure providers, and the telecommunications carriers that supply numbers and carry traffic. That list is updated there first.

Read the Privacy Policy and Terms of Service. To report a security concern or request a security review, use the contact form. For commercial questions about deployment, use the sales form.