AI Disclosures
Last updated May 5, 2026 · Version 2026-05-05
In plain English
- Juno's briefings — both the text and the spoken audio — are generated by AI.
- We send your topics, prompt, and feedback to OpenAI and (depending on configuration) DeepSeek to write the briefing. We use OpenAI's in-tool web search or Exa to find the underlying news articles.
- We send the resulting script to OpenAI's text-to-speech (or, depending on configuration, Inworld or Lemonfox) to produce the audio you hear.
- We do not send your phone number, email, password, or payment information to any AI provider.
- AI is fallible. Always verify a Juno briefing before relying on it for important decisions.
What in Juno is AI-generated?
The following content in Juno is generated by AI and is marked as such here for transparency under the EU AI Act and for parity with equivalent transparency obligations elsewhere:
- Your personalized prompt (the natural-language description of the briefings you want), if you let Juno write or refine it for you.
- The article summaries in each briefing.
- The show plan (the editorial ordering and framing of stories).
- The voice script (the text that is spoken on your call or shown as a transcript).
- The spoken audio on your daily call. The voice you hear is a synthesized voice produced by a text-to-speech model, not a recorded human reading.
Headlines, source URLs, and publication metadata are drawn from third-party news sources and are not AI-generated, although their selection (which stories to include) is.
AI providers we use, what data we send each one
The exact model selected at runtime can vary as we tune quality; the names below reflect the current default configuration. The environment variable column is provided so engineers reviewing this page can match it to the server configuration in apps/server/src/lib/env.ts.
OpenAI
Role: Primary provider for chat-completion-style writing tasks (prompt generation, show planning, article summarization, voice script writing) and the default text-to-speech provider. Also performs in-tool web search on the topics you select.
| Surface | Default model | Data we send |
|---|---|---|
| Onboarding briefing + brief generation | gpt-5-mini | Your topics, goal, angle, tone, schedule, "always cover" text, and current briefing-feedback entries. The model produces a JSON show plan and search queries. |
| Article summarization | gpt-5-nano | Article titles, sources, descriptions, and short excerpts that came back from the search step. |
| Voice script writing | gpt-5-mini | The summarized story briefs plus your prompt, tone, and recent feedback. |
| Text-to-speech | gpt-4o-mini-tts | The completed voice script and voice / pacing settings. |
DeepSeek
Role: Alternate / fallback provider for the same chat-completion writing tasks above. Used when the summarization or showrunner provider is configured to DeepSeek, or when the OpenAI path returns an error.
| Surface | Default model | Data we send |
|---|---|---|
| Summarization, show plan, voice script | deepseek-v4-pro | The same data as the OpenAI equivalents above, when the DeepSeek path is selected. |
Exa
Role: Optional web-search provider, used when WEB_SEARCH_PROVIDER=exa is configured. We send a search query derived from your prompt; Exa returns titles, URLs, and snippets, which are then summarized by OpenAI or DeepSeek.
Inworld
Role: Optional text-to-speech provider, used when TTS_PROVIDER=inworld is configured. We send the voice script and voice configuration. Default model: inworld-tts-1-max.
Lemonfox
Role: Optional text-to-speech provider, used when TTS_PROVIDER=lemonfox is configured. We send the voice script and voice configuration.
What we never send to AI providers
- Your phone number.
- Your email address.
- Your password.
- Payment information.
- The audio of your phone call. Calls are one-way playback.
- Your contact list or device identifiers.
Training
We do not authorize our AI providers to use your inputs to train their models. Each provider has its own retention and training policies; we direct you to their published terms for the authoritative statement (OpenAI, DeepSeek, Exa, Inworld, Lemonfox).
Accuracy and the limits of AI
AI-generated content can be wrong, even confidently wrong. A Juno briefing can:
- Misattribute a quote, a number, or a fact to the wrong source.
- Summarize an article in a way that loses important nuance.
- Conflate two different stories, or invent a story that does not exist (a "hallucination").
- Reflect bias present in the underlying training data.
Treat a Juno briefing as a starting point for your own research, not as a substitute for primary sources. We always cite the articles a briefing draws from; click through and verify before you act on anything important. Do not rely on Juno for medical, legal, financial, or safety-critical decisions.
Human oversight
The Juno team monitors briefings produced by the pipeline, reviews user feedback, and continuously adjusts prompts and provider configuration to improve quality. We do not, however, review every briefing before it is delivered — the volume makes that infeasible. Your feedback (the "Tune Juno" sheet in the app) is the primary mechanism for correcting recurring problems with your specific briefing.
EU AI Act notice
Juno produces synthetic audio and synthetic text (Article 50(2) and 50(4) of Regulation (EU) 2024/1689). This page constitutes the disclosure that a user is interacting with an AI system and that the output is artificially generated. The corresponding machine-readable disclosure is included in our briefing audio metadata; details on request.
We are continuing to monitor implementing acts and codes of practice and will update this page as obligations come into force.
Contact
Questions, concerns, or AI-quality reports: email privacy@junoai.app.