Voice AI Disclosure

Last updated: 2026-05-14 · These terms are reviewed by Astra periodically and effective on the Last updated date above.

This disclosure applies to every voice session on Astra, including AI voice readings, voice-cloned personas, and live voice / video calls with a reader. Read alongside the Astrology & Divination Disclaimer, Privacy Policy, and (for paying users) Subscription Terms.

1. AI is AI, human is human

When you tap "Talk to your reader" you may be connected to:

an AI reader — a synthetic voice powered by large-language- model output (the default for Voice readings); or
a human reader — a verified independent practitioner on the marketplace.

You will always know which:

AI readers carry a clear "AI" label and a robot badge in the room header.
Human readers carry a verified handle and avatar.
Human readers are contractually required to answer honestly if you ask "are you AI or a person?" Misrepresenting AI as a human is a ban-level violation.
The AI is built to answer "I'm an AI" truthfully.

2. How the AI voice is produced

Our voice pipeline composes three independent services:

Anthropic Claude — generates the response text from your question and context.
Deepgram — transcribes your speech to text.
Cartesia — turns the response text into the synthetic voice you hear.

Each is independently contracted; see the Privacy Policy sub- processor list.

3. Voice clones and likeness rights

Some voice personas are clones built from a real person who has licensed their voice to Astra in writing under a recorded- consent contract. The persona is not a live recording of that person and the person is not speaking to you in real time.

We do not publish voice clones built from third parties without written consent and we do not allow users to clone non- consenting voices on the Platform. If you believe a persona uses your voice without permission, email hello@astraplatform.ai.

Voice readings (1:1 with the AI) — by default we do not retain the audio after the session. We may keep a 30-day metadata log (start, duration, persona) for trust & safety. You may opt in to Transcript history in Settings to retain text transcripts.
Reader voice sessions (1:1 with a human reader) — sessions may be recorded for safety, quality, and dispute resolution. You and the reader are notified at session start with an explicit consent prompt. If either party declines, recording does not happen.
Live voice / video broadcasts — may be recorded as VOD by the broadcaster. Recording is disclosed in the room header.
Retention: reader-session recordings and VOD are retained for 30 days by default. You may request earlier deletion at any time via Settings → Privacy → Recordings.
Two-party-consent jurisdictions (California, Florida, Pennsylvania, and others) — by joining a session that displays the "recording" indicator, you consent to the recording. If you do not consent, leave the session.

5. What the AI cannot do

It is not a licensed therapist, physician, psychiatrist, lawyer, accountant, financial advisor, or clergyperson, and it is not acting as one.
It cannot diagnose, prescribe, forecast, or predict outcomes.
It cannot make commitments on your behalf.
It does not remember you between sessions unless you sign in and explicitly enable the Journal feature.
It may produce inaccurate, biased, or hallucinated output. Treat it as a thoughtful friend reading symbols, not a source of fact.

6. Safe use

Use AI voice for self-reflection, journaling, brainstorming, language practice, or exploring symbolism.
Do not rely on AI voice for medical, psychiatric, legal, financial, or relationship decisions. Speak to a qualified professional.
If you are in crisis, the AI will pivot to crisis-support scripting and surface the hotline numbers below. Please call a real human.

7. Crisis carve-out

Astra is not a crisis service. If you are in danger, contact emergency services immediately.

Always-available hotlines (full list in Crisis & Safety):

US: 988 Suicide & Crisis Lifeline (call or text)
UK / ROI: 116 123 Samaritans
Australia: 13 11 14 Lifeline
Canada: 988 Suicide Crisis Helpline
Mainland China: 400-161-9995 Hope24
Hong Kong: 2896 0000 Samaritans HK
Taiwan: 1925 安心專線
Japan: TELL Japan 03-5774-0992
Korea: 1393 자살예방상담전화
Germany: 0800 111 0 111 Telefonseelsorge
France: 3114
Brazil: 188 CVV
India: AASRA +91-9820466726
Global directory: https://findahelpline.com

8. Data handling for voice

Audio in transit — TLS 1.3 encrypted to Deepgram and Cartesia.
Audio at rest — not persisted by default. Reader-session recordings stored encrypted in Supabase Storage and accessible only to participants and the Astra trust-and-safety team on a documented basis.
Sub-processors — see the Privacy Policy. Anthropic, Deepgram, and Cartesia are contractually prohibited from training general models on Astra audio inputs.
You may request deletion of any retained transcript or recording at any time via Settings → Privacy → Recordings or by emailing hello@astraplatform.ai.

9. Right to a human

If at any time you would rather talk to a human reader, end the AI session and book one from /readers. We will not penalise you.

10. Reporting AI behaviour

If the AI produces output that violates our Community Guidelines (harassment, sexual content, self-harm encouragement) or this disclosure, please report it via the in-room Report button or email hello@astraplatform.ai. We review every report and adjust safety prompts accordingly.