COPPA Direct Notice
Effective Date: 2026-11-01 (pilot release; effective when this version is deployed) Policy Version: 2026-11-01 2026-07-17: product name updated from Nova to Lantern throughout (branding alignment; the same product, practices unchanged), and transcript retention simplified to the operating mechanism (deleted on your deletion request); no substantive change. 2026-10-21: the company that runs Lantern is now Lantern Learning Labs, Inc., a Delaware corporation (previously Scout Learning Lab LLC); updated our mailing address and phone number; no change to what we collect from children or how we use or share it.
This is the "Direct Notice" required by the FTC's COPPA Rule (16 CFR §312.4(c)) — a clear, complete statement that we provide directly to parents before we ask for their consent.
2026-11-01 pilot update: browser-based account, consent and child setup; signed-name evidence retained; a separate, clearly disclosed forensic-recording choice for invited pilot families. This choice is selected initially, can be declined before signing, and is not a grant until you submit your signed decision.
1. Who we are
Lantern Learning Labs, Inc. 3178 Abington Drive, Beverly Hills, CA 90210 · privacy@lanternlearning.io · (310) 428-0424
We make Lantern, an AI tutor that helps children practice reading, writing, and math. We are the operator under COPPA.
2. What we collect from your child
- First name and age; optional grade, pronouns, interests, reading starting point and everyday learning supports. The pilot setup form does not ask for an exact birth date or diagnosis notes
- Voice audio during sessions (streamed to OpenAI for the spoken conversation and to Deepgram for reading transcription — see §3)
- Camera frames of the page or worksheet your child shows the device (uploaded to Google for text recognition; a short-lived copy stays on the device up to 7 days)
- Text transcripts of what Lantern hears (stored on your device, with a separate cloud copy only if you submit the pilot forensic-recording consent below; shared transiently with Anthropic for the post-session analysis described in §5; deleted on your deletion request)
- Books, worksheets, and handwritten work shown to Lantern
- A unique device identifier so the app works
- Notes Lantern writes about how the session went, including a weekly progress summary written for you
We do NOT collect: your child's last name, home address, phone number, school name, social media accounts, or any biometric identifier (voiceprint, face embedding, fingerprint).
3. What happens to your child's voice
Your child's spoken audio streams live to OpenAI's Realtime API (model gpt-realtime-2) to power the back-and-forth conversation, and separately to Deepgram to convert speech to text for the reading assessment. Neither provider uses your child's voice to train AI. OpenAI keeps voice inputs and outputs for up to thirty (30) days for abuse monitoring and then deletes them (its Realtime service also holds up to about one (1) hour of audio to keep a multi-turn conversation going); Deepgram does not keep the audio after transcribing it. We do not store raw voice audio in our own cloud unless you separately consent to the pilot forensic-recording copy described below. We never create a voiceprint of your child. (Google's Gemini Live is our standby conversational provider; if we ever switch to it, your child's voice would stream to Google under the same commitments.)
Research-cohort exception: if you opt in to our research cohort, a copy of your child's reading-session audio stays on your local device — never our cloud — to help us improve Lantern's reading help. You can turn off the cohort opt-in at any time from the consent dashboard at https://consent.lanternlearning.io; withdrawing files a deletion request that our team completes within thirty (30) days, with email confirmation.
3.1 Separate pilot forensic-recording choice
For invited friends-and-family pilots, we offer a separate recording choice to help authorized Lantern Learning Labs staff diagnose problems and improve tutoring quality and safety. It is selected when the consent page first opens, is clearly labeled, and you can turn it off before signing. Opening the page does not grant recording permission. Your submitted signed choice applies to the child profiles explicitly covered in that setup or renewal.
If you consent, the device may upload a diagnostic session bundle to our private Supabase database and Storage: verbatim transcripts, reading-session audio when available, still images of pages/worksheets captured during the session, assessment and event details, logs, and system/configuration snapshots. These are still page images, not a continuous video recording. Access is limited to authorized staff operating the pilot; the parent setup website does not expose the private recordings. This consent does not authorize using the bundles to train AI models.
The configured retention period is thirty (30) days from upload. Scheduled cleanup processes expired bundles; failures remain tracked for retry. You can withdraw at the setup website or email privacy@lanternlearning.io. Withdrawal disables the child's forensic-upload permission and creates a request to remove retained bundles. Already-uploading or already-uploaded data is handled through that tracked removal process; a pending request is not a claim that every copy has already been removed. Other retention commitments below continue to apply to ordinary tutoring data and consent records.
Forensic recording is separate from the optional research-cohort/on-device audio-retention choice and from account-email verification. A declined recording choice does not become a recording grant just because core tutoring consent is accepted. Existing families are not opted in automatically.
4. What stays on your device vs. goes to our cloud
| Stays on your device | Goes to our cloud or a provider |
|---|---|
| Full transcript of every session | Your child's first name and grade; a forensic transcript copy only with separate pilot consent |
| Short-lived copy of camera frames (≤ 7 days) | Camera frames + handwritten work images → Google for text recognition (OCR) |
| Copy of handwritten work images (no automatic cleanup yet — removed when you request deletion) | Voice audio → OpenAI (conversation) and Deepgram (reading transcription) |
| Short reading-practice recording (deleted within a day or two) | AI-generated session summaries + weekly progress summaries, reading scores, words learned, books read |
| (Research cohort: reading audio) |
Outside a separately consented pilot forensic bundle, your child's voice is not stored in our own cloud, but it is processed by OpenAI and Deepgram as described in §3 — OpenAI keeps voice inputs and outputs up to thirty (30) days for abuse monitoring, then deletes them. During reading practice the device also keeps a short local recording, deleted automatically within a day or two. Camera frames and handwritten work images are uploaded to Google for text recognition (OCR); a copy stays on the device as shown above.
5. Who else sees the data
Five core service providers:
- OpenAI — real-time spoken conversation (your child's voice). No training on API data; inputs and outputs kept up to 30 days for abuse monitoring, then deleted.
- Deepgram — speech-to-text for the reading assessment. Audio not retained after transcription.
- Google — text recognition (OCR) on the camera images of pages and handwritten work; also our standby conversational provider (Gemini Live — your child's voice would stream to Google only if we switch to that standby). No training on input.
- Anthropic (Claude) — session analysis. No training on input.
- Supabase — cloud database for summaries and account info, signed consent evidence, and separately consented pilot forensic bundles in private Storage.
A complete subprocessor list is at subprocessors.md.
None of these providers use your child's data to train AI, for advertising, or for any purpose beyond providing the service to us.
6. How long we keep things
| Item | Retention |
|---|---|
| Voice (conversation + transcription) | Outside separately consented forensic bundles, not stored in our cloud; processed by OpenAI (kept up to 30 days for abuse monitoring, then deleted) and Deepgram (not retained after transcription) |
| Reading-practice recording (on your device) | Deleted automatically within a day or two |
| Reading audio (research cohort only) | Kept while in the cohort; withdraw anytime (removed within 30 days) |
| Camera frames | ≤ 7 days |
| Transcripts (on your device) | Deleted on your deletion request |
| Cloud summaries (session + weekly) | ≤ 730 days (target — automatic age-out planned; sooner on request) |
| Profile and account | Until you delete |
| Consent records, including the pilot signer name and submitted recording choice | 7 years under our consent-evidence retention policy |
| Separately consented forensic bundles | 30-day retention setting from upload, processed by scheduled cleanup; earlier withdrawal/deletion requests are tracked until fulfilled |
7. Your rights as the parent
You can:
- See and export the data this service holds about your family as a JSON + CSV bundle (consent dashboard → Your data → Generate export, where available; pilot families may also email privacy@lanternlearning.io)
- Correct any of it (email privacy@lanternlearning.io — we correct within seven (7) days)
- Delete any or all of it (consent dashboard → Your data → Delete, where available, or email privacy@lanternlearning.io; our team completes deletion within thirty (30) days, with email confirmation)
- Pause your child's profile at any time (existing consent dashboard → Your data → Pause, or email privacy@lanternlearning.io)
- Withdraw voice-retention consent at any time (existing consent dashboard → Your consent → toggle off; pilot forensic consent has its own withdrawal action)
- Email privacy@lanternlearning.io with questions; we respond within seven (7) days for COPPA-related requests
8. AI disclosure
Lantern talks to your child using OpenAI's real-time voice model (gpt-realtime-2) for the spoken conversation; Anthropic's Claude analyzes each session afterward; and Google performs text recognition (OCR) on the page and handwriting images. None of these providers train their models on your child's sessions. Lantern's responses are generated fresh each session and are not stored long-term by these providers beyond OpenAI's up-to-30-day abuse-monitoring window (your device keeps the session transcript, as described in §2). We never derive a voiceprint of your child.
Full AI disclosure: ai-disclosure.md.
9. The permission we need from you
By completing the consent wizard at https://consent.lanternlearning.io, you give Lantern Learning Labs permission to:
- Collect the information described above
- Use it to operate Lantern for your child
- Share the limited pieces described in §5 with our service providers
We will not collect, use, or share your child's information for any other purpose without obtaining renewed consent from you.
10. If you change your mind
For the browser pilot, sign in at https://consent.lanternlearning.io to review your saved setup and withdraw forensic recording. Email privacy@lanternlearning.io to request a copy, correction, pause or deletion of your family's data. Existing account-data tools remain available where offered. We acknowledge COPPA-specific deletion requests within seven (7) days; our team then completes deletion of our cloud systems within thirty (30) days and confirms by email. Data on the Lantern device itself is erased as part of that same process, or when the device is returned and re-flashed — reset instructions come with the confirmation email if you keep the device. Non-COPPA privacy requests are acknowledged within ten (10) business days and completed within thirty (30) days.
You can also email privacy@lanternlearning.io with any request.
If you do not return a signed consent form within thirty (30) days of receiving our welcome packet, we will delete the email address and contact information we collected from you, as required by 16 CFR §312.4(c)(1)(vi).
11. How we keep this safe
Data is encrypted in transit (TLS 1.3) and at rest (AES-256). Our service providers are SOC 2 Type II certified. Access to production data is restricted to a small number of authorized Lantern Learning Labs engineers, with audit logging.
Full security disclosure: security.md.
12. Children's Privacy Policy
For the full Children's Privacy Policy, see childrens-privacy.md.
13. Questions or complaints
Email privacy@lanternlearning.io for direct questions. To file an independent complaint, contact:
Lantern Learning Labs is not currently a member of an FTC-approved COPPA Safe Harbor program.
You may also file a complaint directly with the Federal Trade Commission at https://reportfraud.ftc.gov or with your state attorney general.
Questions? Email privacy@lanternlearning.io. To file a complaint, contact the FTC at reportfraud.ftc.gov or your state attorney general.