AI & Educational Content Disclaimer

Version 1.0 — Draft for legal review · Effective: Not yet set · Last updated: 2026-08-29

# SomaSkul AI & Educational Content Disclaimer

Version 1.0 — Draft for legal review Last updated: 2026-08-29

Update 2026-08-31 — Gemini deferred for this launch. GEMINI_API_KEY has been removed from production, so none of the Gemini data flows described below currently occur — ZANA, homework solving, flashcard generation, and Learner Performance Analysis currently degrade to a local/offline response or a clean "not configured" state rather than calling Gemini. This document remains an accurate description of what happens once Gemini is reintroduced, and the §6 billing-tier decision must still be resolved before that happens — but it is not a blocker for the current launch. See docs/legal/GEMINI_PROVIDER_VERIFICATION.md and docs/legal/FINAL_LAUNCH_READINESS_MATRIX.md for the current status.

## 1. SomaSkul uses artificial intelligence

SomaSkul's tutor ZANA, its homework-photo solving feature, its flashcard generation, and its Learner Performance Analysis feature are all powered by AI — specifically Google's Gemini API (see THIRD_PARTY_DATA_PROCESSORS.md). If SomaSkul's optional Research Engine is enabled, ZANA may also draw on live web search results via Tavily.

## 2. AI responses may contain errors

AI-generated content — including step-by-step solutions, explanations, quiz questions, and extracted report-card data — can be incorrect, incomplete, or misleading, even when it sounds confident. SomaSkul's engineering includes a verification pass on the homework-solving pipeline (an independent check of the answer before it's shown to you), and ZANA's self-learning notes require a high-confidence extraction plus a separate verification pass before being saved — but no automated check eliminates the possibility of error.

Always cross-check important academic answers — especially anything that will be submitted for grading, used to study for a national examination, or relied on for a high-stakes decision — against your teacher, textbook, or official curriculum materials.

## 3. ZANA is designed to teach, not to just give answers

By default, ZANA follows a "guided" teaching style: it will not simply hand you a final answer to a question you haven't attempted. This is intentional — it's core to how SomaSkul is designed to help you actually learn, not just complete an assignment. ZANA has other modes (explanation, quiz, exam-prep, and a supportive mode for when you're struggling emotionally, not academically) that behave differently, but none of them is a substitute for the judgment of a real teacher.

## 4. AI does not replace your teachers, parents, or school

SomaSkul is a study aid. It does not replace instruction from a qualified teacher, guidance from a parent or guardian, or your school's own curriculum materials and examination guidance. Do not rely on SomaSkul as your sole source of preparation for a national examination.

## 5. How your input is processed

When you ask ZANA a question or submit a photo, that content is sent to Google's Gemini API to generate a response, as described in PRIVACY_POLICY.md §6-7. A short excerpt of some questions may be retained on SomaSkul's servers as part of how ZANA improves over time — see PRIVACY_POLICY.md §7 and DATA_RETENTION_POLICY.md for specifics. Full conversation transcripts are not uploaded to SomaSkul's servers; they stay on your device.

Report cards and homework pages are images, and images can show your name. SomaSkul no longer includes your name in ZANA's written instructions to Google (as of 2026-08-28), but if you photograph a document that has your name printed on it — most report cards do — that name is still visible to Google as part of the picture itself. SomaSkul does not detect, crop, or blur names in submitted photos. See docs/legal/GEMINI_PROVIDER_VERIFICATION.md §4.

## 6. Third-party AI providers

Verified 2026-08-29 against Google's current Gemini API terms — this materially depends on a billing setting, not on SomaSkul's code:

If SomaSkul's Gemini API usage is on the paid tier, Google's terms state prompts

and responses are not used to improve Google's products, and are logged only briefly for abuse detection.

If SomaSkul's Gemini API usage is on the free tier, Google's own terms state

submitted content *is* used to improve Google's products, may be read by human reviewers, and Google's terms explicitly instruct developers: *"Do not submit sensitive, confidential, or personal information to the Unpaid Services."* [BUSINESS DECISION REQUIRED — urgent]: SomaSkul's operator must confirm which tier production traffic is actually billed under before this policy can state a confirmed position. See docs/legal/THIRD_PARTY_DATA_PROCESSORS.md for the full finding, and docs/legal/GEMINI_PROVIDER_VERIFICATION.md §7-9 for the exact pending verification steps. A boot-time guard (GEMINI_ENVIRONMENT, backend config/env.js) now refuses to start a production deployment unless an operator has explicitly attested the key belongs to a paid-tier project — but that attestation is configuration, not proof; it does not, by itself, satisfy this business decision.

Tavily (used only if the optional Research Engine is enabled) publishes a privacy

policy claiming "zero data retention," but a secondary review flagged some internal inconsistency between its retention and data-rights sections — EXTERNAL PROVIDER VERIFICATION REQUIRED: read Tavily's current policy directly before relying on this.

Do not publish a claim of "your data is never used for training" until the Gemini billing tier is confirmed and, if necessary, changed.

*This document is Version 1.0 and has not yet been reviewed by legal counsel or published. It is incorporated by reference into TERMS_OF_SERVICE.md §8.*