Security

Architecture

All processing and storage takes place on United States infrastructure. Data at rest is encrypted with AES-256 through Azure Transparent Data Encryption; data in transit uses TLS 1.2 or higher. Secrets and keys are held in a managed, HSM-backed key vault with access logging and rotation, and backups are geo-redundant within US regions and encrypted under the same scheme.

Every request passes through validation, role-based access control and input sanitisation before it reaches an application service. Edge protection includes a web application firewall with OWASP-aligned rules and rate limiting.

HIPAA

For healthcare and other sensitive deployments, calls and records are routed through a dedicated HIPAA-grade pathway. The protections below are enforced automatically for those deployments.

• Zero AI training. No voice data, transcript, document or message is ever used to train any model.
• US storage only. Records are stored exclusively in Microsoft Azure SQL in US data centres.
• BAA coverage end to end. Every provider in the data path is covered by a signed Business Associate Agreement — speech recognition, the model, synthesis, analysis and storage.
• No third-party recording. Carrier-side recording is disabled for HIPAA-grade calls; recordings are retained only within the platform, and only to deliver the service.
• Complete audit trail. Access to and modification of records is logged, including requests that policy refused.

Compliance enforced at runtime

Controls that exist only in a policy document do not survive contact with production. Pendral enforces them while the work is happening: numbers are scrubbed against the federal Do Not Call registry, state lists and your own suppression list before a call is placed, and opt-out requests — explicit or implied — are detected and honoured during the call itself. Consent type is tracked for TCPA purposes, AI disclosure is offered on request, and an expanded set of opt-out keywords is actioned immediately.

Safety escalation is built into every voice agent. Signals of crisis, severe distress or medical emergency route to qualified human support with a callback rather than a disconnection, and every escalation is logged for review.

These features provide compliance assistance, not a guarantee. Legal responsibility remains with the customer.

Identity and access

One permission model spans all four products: organisation, team, project, environment and resource, with a small set of named roles resolved consistently everywhere. Cloud access is brokered per operation rather than stored, so there is no long-lived credential to leak or rotate, and an AI agent acting on your infrastructure is always scoped strictly below the person who asked for it.

Access is governed by enterprise single sign-on, passwordless login and validated API keys. Data classification travels with the record and constrains which models, tools and members may touch it.

Sub-processors

Pendral maintains signed Business Associate Agreements with every provider that handles protected health information across the full stack. This gives customers a single, consistent point of compliance rather than a chain of separate agreements to negotiate.

The current sub-processor list, along with our BAA and Data Processing Addendum, is available to customers and to prospective customers under evaluation. Request it from austin.awes@pendral.io.

Reporting a vulnerability

If you believe you have found a security issue in a Pendral product, contact austin.awes@pendral.io with enough detail to reproduce it. We will acknowledge the report and keep you updated through remediation. Please do not disclose publicly until we have had a chance to fix it.