AI Transparency Statement

Effective date: July 7, 2026 · Last updated: August 1, 2026

Sona builds digital twins: AI models of real people, created by those people themselves. Because our product speaks in your voice and style, clear labeling matters. This statement explains how Sona identifies AI output and obtains consent. It also describes the system's capabilities and limits.

1. Always labeled as AI

• Every Sona twin is clearly identified as an AI. In any conversation with a twin, the interface displays that you are talking to a digital twin, not a live human.
• Signed Persona: supported output formats carry a cryptographic signature that identifies the source as a Sona twin. The interface also displays an AI disclosure.
• When a twin communicates on your behalf with third parties (for example, drafting or attending a meeting), the recipient is informed that they are interacting with an AI twin ("Sona on" disclosure).

2. What a Sona twin is

A Sona twin is a statistical AI model that can imitate documented communication patterns. It has no consciousness or feelings, and the product must clearly identify it as AI.

3. Consent and permissions

• A twin can only be created by the person it models, about themselves, with their explicit consent.
• Every expansion of what a twin can do (voice cloning, acting on your behalf, access by others) requires a separate, revocable consent.
• Consent settings and permissions are recorded in the twin's rights registry and can be reviewed and revoked at any time.
• We prohibit and technically counter the creation of twins of third parties without their consent (identity verification, abuse detection, signature enforcement).

4. Capabilities and limits

What twins do well: conversation in your style, retrieval of your documented memories, drafting, summarizing, advising within your recorded context.

Known limits, stated honestly:

• Twins can make mistakes. Generative AI can produce plausible but wrong output. Truth Anchor constrains your twin to documented facts about you and instructs it to say "let me check with the real me" rather than invent. No safeguard is perfect, so review anything important.
• Twins are not professional advisors. Their output is not medical, legal, financial or psychological advice.
• Twins reflect their material. A twin knows only what you chose to teach it; gaps in capture mean gaps in the twin.

5. Human oversight and control

• The twin asks before it acts, within a mandate you configure; every action is logged and auditable.
• You can correct the twin's facts, constrain its topics (Values Firewall), pause it, or delete it entirely at any time.
• We design against unhealthy dependency: the twin encourages contact with real people and does not present itself as a substitute for them.

6. How our models are built

Sona combines its personalization layer with the models and service providers described in current technical documentation. Personal data is processed for the purposes stated in the Privacy Policy. Our fidelity metric, the blind recognition test, is measured only with the explicit participation and consent of everyone involved.

7. Regulatory alignment

We align with the EU AI Act's transparency requirements for AI systems that interact with humans and generate synthetic content (including labeling of AI-generated output), with GDPR and with California's CCPA (see our GDPR, UK GDPR & CCPA Statement), and we monitor emerging rules on digital likeness and post-mortem data across the jurisdictions where we operate.

8. Questions, reports, disclosures

If you find unlabeled twin content, suspect impersonation, or have a transparency question, contact ai-transparency@sonatwin.com. We publish material updates to this statement.