AI Disclosure

Effective Date: 2026-07-16 Last Updated: 2026-07-16

What this page is

Lantern uses generative AI (large language models, or "LLMs") to talk with your child and assess their reading and writing. This page tells you, in plain English:

1. Which AI models we use
2. What they receive from your child's sessions
3. Whether those models train on your child's data
4. What we do to keep their output appropriate for children
5. Your rights regarding AI processing

1. Which AI models we use

We may add or change models over time. Material model changes are reflected in updates to this page and announced via the Parent App.

2. What each model receives

OpenAI (real-time conversation — the production provider):

• Voice audio streamed during a session (not stored in our cloud; OpenAI keeps inputs and outputs up to 30 days for abuse monitoring, then deletes them)
• System prompt instructions
• Brief conversation context

Google (text recognition + standby conversation):

• Camera frames of book pages and handwriting, sent as images for OCR (transient — not retained beyond the on-device copy)
• If the Gemini Live standby provider is active: voice audio streamed during a session (transient), system prompt instructions, brief conversation context

Deepgram (speech-to-text):

• Voice audio (transient — deleted after transcription)
• Custom vocabulary (book titles, child's first name) for accuracy

Anthropic Claude (post-session analysis + weekly summaries):

• Text transcript of the session
• Aggregated metrics (reading accuracy, words attempted, error patterns)
• Reference to prior session summaries for continuity
• System prompt with analysis instructions

What each model does NOT receive:

• Your last name, address, phone, school
• Your child's full name (only first name)
• Your child's biometric identifiers
• Long-term identity tokens
• Payment information

3. Training and retention by the AI providers

We have contractual commitments from each provider:

• None of these providers train AI models on the data we send them.
• Providers do not retain inputs beyond the operational window needed to deliver the response (for OpenAI that window includes an abuse-monitoring copy kept up to thirty (30) days, then deleted).
• Providers do not use inputs for advertising, profile-building, or any purpose other than delivering the requested AI completion to Scout.

We confirm these terms at each subprocessor onboarding and at annual review. The relevant provider terms are their published data-processing, usage, and retention policies:

• OpenAI (Realtime API) — data processing addendum, API no-train terms
• Google (Gemini OCR + Gemini Live) — data processing terms
• Anthropic (Claude) — usage policy
• Deepgram — data retention terms

4. Safety guardrails

Before any AI response reaches your child:

1. System prompt instructs the model to follow age-appropriate, child-safe behavior
2. Content filtering removes responses that fall into blocked categories
3. Topic detection routes sensitive topics to "talk to a trusted adult" redirects
4. Provider-level safety systems run before our application-level filters

Full safety details: safety.md.

5. Voiceprints and biometric identifiers

We never create:

• A voiceprint of your child
• A face embedding or face match identifier
• Any other biometric identifier from your child's interactions

Voice audio is processed in the moment — for speech recognition and for the real-time conversation — and is never stored in our cloud (during the friends-and-family pilot, a short reading-practice recording stays on the family device for up to a day or two). The text transcript is the only persistent representation, and even that lives only on your local device (cloud only sees aggregated summaries).

This commitment is at the contractual level with our providers (none of them are permitted to generate biometric identifiers from data we send) and at the architectural level (the daemon code does not request biometric features from any API endpoint).

6. AI limitations and known issues

AI is not perfect. We try to be honest about this:

• AI can be inaccurate. Educational content from Lantern is meant to help your child practice, not to replace authoritative sources. For high-stakes questions, verify with a teacher, parent, or trusted reference.
• AI can be inappropriate occasionally. Our safety guardrails are strong but not perfect. If Lantern produces something inappropriate, please report it (safety@lanternlearning.io) so we can investigate.
• AI is not a substitute for human supervision. Lantern is a learning tool, not a babysitter, therapist, doctor, or counselor.
• AI does not understand your child. Lantern adapts to patterns over time, but it does not have a true understanding of your child's emotional state, family context, or developmental specifics.

7. Your rights

You have the right to:

• Know that AI is processing your child's data (this page)
• See the input and output of any AI processing (Parent App → Session transcripts)
• Opt out of AI features that are not core to Lantern (TBD — depends on which features are optional)
• Withdraw consent at any time (Parent App → Consent → toggle off)
• Object to processing for any specific purpose (email privacy@lanternlearning.io)

For California residents, additional AI-specific rights may apply under CCPA/CPRA — see your-privacy-choices.md.

8. Changes to AI providers and models

When we add, remove, or substantially change AI providers or models:

1. Update this page with the new effective date
2. Notify Parent App users at least thirty (30) days before the change takes effect (sooner if required by safety or legal urgency)
3. Update the Subprocessor List
4. If the change constitutes a material change in practices under COPPA, trigger re-consent

9. Disclosure of AI use to children

Lantern introduces itself as an AI tutor in age-appropriate language. We believe children deserve to know they're talking to an AI, not a person. Lantern does not pretend to be a human.

10. Independent oversight

Scout is not currently a member of an FTC-approved Safe Harbor program. If we join one in the future, that program will independently audit our AI practices and provide an additional dispute resolution path.

Contact

ai-questions@lanternlearning.io privacy@lanternlearning.io safety@lanternlearning.io

Questions? Email privacy@lanternlearning.io. To file a complaint, contact the FTC at reportfraud.ftc.gov or your state attorney general.