Radio Latte AI and Catalog Transparency Statement
Effective date: 2026-05-21 Last updated: 2026-06-25
This statement applies to both products described in our Terms of Service:
- The Venue Service catalog used by business venues.
- The Mobile App (iOS), specifically the Creators flow that helps individual consumers turn lyric text into a song.
It explains how we use AI, what happens with content you submit, and what we do to keep harmful content out of the service.
A. Short Version
- Songs in the Venue catalog and songs created through the Mobile App’s Creators flow may be produced with AI assistance.
- Radio Latte controls the catalog. We decide what is and is not included.
- In the Mobile App, your lyric text is processed by AI safety and structuring services. It is not stored on our servers after the AI call returns.
- We do not use your lyrics to train AI models. Our AI subprocessors do not use your lyrics to train their models under the contractual terms we have selected.
- We run multiple safety layers before any song is generated, plus a community-reporting layer for songs already in the catalog.
B. AI in the Venue Catalog
This section continues to apply to the Venue Service.
B.1 How catalog music is created
Radio Latte catalog candidates may be created using a mix of internal composition systems, human creative direction, AI-assisted generation tools, audio processing, quality checks, and human review.
Before music is approved for the Radio Latte catalog, we aim to check that it:
- was created from an approved source;
- does not use customer-uploaded music;
- does not use commercial song samples;
- was not prompted with artist names, song names, labels, film scores, TV themes, or other protected works;
- is reviewed for venue fit and originality concerns;
- is approved for controlled streaming through Radio Latte.
B.2 Production tools (catalog)
Some catalog candidates may be created with assistance from internal or open-source AI music generation tools. These tools are technology inputs. They are not represented as Radio Latte partners, sponsors, licensors, or endorsers unless we enter into separate written agreements.
Radio Latte remains responsible for deciding what enters the Radio Latte catalog.
B.3 What Venue customers receive
Venue customers do not receive track ownership or downloads. Customers receive a limited venue license to play the Radio Latte catalog through Radio Latte at the named venue while their subscription is active.
B.4 Track-level technical records
Rightsholders, rights representatives, licensing reviewers, legal advisers, and diligence reviewers may request reviewed access to track-level technical records at radiolatte.com/compliance. Approved access may include track identity, public creator name, creation or review status, AI-use stamps, lyrics-source stamps, prompt provenance, and prompt exclusion records where stored.
For newer tracks, prompt provenance may be an exact bake-time record. For older tracks, provenance may be reconstructed from available ticket and catalog records, or may be unavailable if no reliable record exists. We label the record state when we disclose it.
We do not provide open public search of these records. We also do not disclose user identity details, secrets, tokens, internal URLs, security-sensitive information, or business-sensitive implementation details through this access path.
C. AI in the Mobile App / Creators Flow
This section applies to the Mobile App on iOS.
C.1 What the Creators flow does
In the Mobile App, you can write lyrics in your own words and have a song generated from them. The flow has five steps; this statement covers steps 1 (writing) and 5 (baking), where AI is involved.
C.2 Where your text is processed
When you tap “Create AI version” (Free style mode) or “Validate” (either mode), Radio Latte processes the text you wrote in one of two paths:
-
On-device (preferred when available). If your iPhone supports Apple Intelligence (Foundation Models), structuring and safety review run on-device. The text you submitted does not leave your device for these steps.
-
Cloud fallback. If on-device AI is not available, the text is sent over HTTPS to Cloudflare Workers AI (Llama 3.1 8B Instruct), hosted by our subprocessor Cloudflare. The text is sent only for the duration of the call. We do not store it on our servers after the response is returned.
C.3 What we send, what we store
| Stage | What we send | Where it goes | Where it is stored after the call |
|---|---|---|---|
| Lyric structuring | Your free-text lyrics and title (if any), language code | Apple Intelligence (on-device) OR Cloudflare Workers AI (cloud) | Not stored on Radio Latte servers. Not stored on your device by us beyond your local draft (which you control via the in-app Discard action). |
| Safety check at Validate | Your title and lyrics combined, language code | Same routing as above | Not stored on Radio Latte servers. |
| Bake (song generation) | An opaque ticket identifier, title, artist name, structured lyric sections, language, and duration | Our music-generation backend (Sicktempo). The backend never receives your username, email, or device identifier — only the opaque ticket. | The generated song is stored in the Radio Latte catalog. The original lyric text is stored only as part of the song’s metadata. |
We do not use your text to train Radio Latte AI models.
Our cloud AI subprocessor (Cloudflare Workers AI) does not use customer prompts to train base models under the contractual terms we have selected. See their own AI policy for the authoritative position.
Apple’s on-device Foundation Models do not transmit your text off-device. See Apple’s published Foundation Models privacy posture.
C.4 The safety pipeline
Radio Latte runs the following safety layers on Mobile App User Content. The system is designed so that an attempt to bypass any one layer is caught by the next.
- Programmatic profanity and slur filter. Lyric text and title are matched against a maintained word/phrase list inside the Mobile App before any AI call. This catches direct slurs, profanity-heavy text, and named-target threats without requiring an external API call.
- Title-only stricter list. Words that are acceptable inside lyric bodies (for example, contextual use of “hate” or “death” as metaphor or theme) are not acceptable as song titles, where they appear without context in lists, lock screens and search. A separate stricter list runs against the title field.
- Apple Intelligence on-device safety review (when available). The on-device model returns a structured verdict on whether the content is appropriate. The check is run at the Validate gate before song generation.
- Cloud AI safety review. When on-device review is unavailable, the Validate gate runs the same review against Cloudflare Workers AI with an explicit content-policy system prompt. If the model cannot run or cannot return a verdict, the gate fails closed and the user cannot proceed.
- AI refusal at the lyric-structuring step. The cloud lyric-structuring endpoint refuses to generate lyrics from text that contains or implies hate speech, sexual content involving minors, glorification of violence, targeted harassment, or promotion of illegal acts. Refusals are surfaced to the user with a reason.
- Bake-side filters. Our music-generation backend (Sicktempo) applies its own pre-bake filters on the structured lyric payload before producing audio.
- Community reporting after publication. Every song in the Mobile App catalog is subject to a user-report mechanism inside the app. Tracks that receive repeated skips or dislikes are automatically returned to the admin queue for re-review (a “skip-to-reverify” rule). Persistent violators of the Acceptable Use Policy are blocked from the Creators flow at the server.
The active in-app block list and the content policy are documented in our Acceptable Use Policy.
C.5 What we do when content is refused
If any safety layer refuses your content, the Mobile App tells you which category was flagged and asks you to edit. We do not save refused content to our servers. We do not penalize you for a refused submission beyond the design of the in-app star economy (a star is spent at the first AI action of a draft, whether the action succeeds or is refused; this is described in the Mobile App and in the Terms of Service).
C.6 Artist credit
Songs created in the Mobile App display the username you chose or accepted as the artist. The song’s title carries the suffix (RL) appended to identify Radio Latte–produced songs (so a song you titled “Super Night Out” renders as Super Night Out (RL)). This stamp is consistent across the App, downloaded files, and the Venue catalog.
We are not the author of the lyric content you submitted. You are. The Terms of Service describe the license you grant us to host, stream and credit the resulting song.
D. What customers receive (both products)
- Venue customers receive a limited venue license to play the Radio Latte catalog through Radio Latte at the named venue while their subscription is active. Venue customers do not receive track ownership or downloads.
- Mobile App users receive free streaming of the catalog (RadioLatte Live!), and the ability to add their own creations to the catalog using stars purchased through the App Store. Mobile App users do not currently receive a downloadable audio file of the songs they create through the Creators flow; this may change in a future release, in which case this statement will be updated.
E. What we do not claim
We do not claim that AI-generated music is automatically copyright-free.
We do not claim that every possible music-rights obligation for a venue disappears.
We do not license music from other sources. If a venue plays music outside Radio Latte, that use is outside the Radio Latte license.
We do not represent the AI tools we use as Radio Latte partners, sponsors, licensors, or endorsers unless we enter into separate written agreements with them.
F. Contact
Questions about how AI is used in Radio Latte: legal@radiolatte.com