befall AI and Data Use Notice
This prominent notice explains when eligible adult interactions may be used for labeling, evaluation, fine-tuning, and training, and how an app-scoped opt-out works.
Default use of eligible adult data
Entry begins at age 16; ages 16–17 use the safety tier, and training eligibility begins at age 18. After the applicable versioned notice is shown, eligible adult conversations, generated content, reactions, ratings, regenerations, edits, report outcomes, and related usage events may be used by default for safety improvement, quality evaluation, labeling, model evaluation, fine-tuning, and training. This is not described as optional opt-in consent, and minor or unknown-age data is never eligible.
Prominent notice before use
The default-use notice must be presented clearly before signup or the first conversation and recorded with its version and display time. Authority never backdates itself: older data is eligible only if covered by a valid legacy opt-in, while new default-use authority covers only events created after the versioned notice was shown.
Operational processing continues after opt-out
A training opt-out does not stop processing needed to generate an AI response, store a conversation or requested memory, perform age and content-safety checks, investigate reports or abuse, secure the service, process payments, provide support, or comply with law. Account deletion, conversation deletion, privacy-rights requests, and training opt-out remain separate operations.
Pseudonymization and dataset separation
Before eligible data enters an improvement workflow, direct identifiers such as email, phone number, account ID, payment identifier, and exact address are removed. Names, contact details, addresses, account references, financial data, and identity-document data in free text are automatically detected and, where needed, manually masked. Training identifiers are separated from operational account IDs, source stores are logically or physically separated, and reidentification keys have separate access controls.
Labeling access and accountability
Only authorized personnel or a provider bound by verified confidentiality, security, access, and deletion duties may see the minimum context needed for a labeling task. No external labeling provider may be used before those terms and public transfer facts are complete. Labeling views must not expose account or contact details, and access, viewing, export, and deletion activity is logged.
Data excluded from improvement
All minor and unknown-age data is excluded, as are actual-user health, diagnosis, sexual-life, sexual-orientation, political, religious, financial, identity-document, or payment-credential data; self-harm, suicide, abuse, exploitation, or emergency crisis conversations; material under a report or rights hold; and post-opt-out events. Adult fictional roleplay is eligible only when it contains no actual user sensitive data and passes safety and rights filters.
What an opt-out changes
An effective opt-out stops later-created data from entering training and long-term improvement retention. The user's data must be removed from pending training queues, uncompleted labeling jobs, and separable datasets, and no new evaluation, fine-tuning, or training task may be created from it. Any copy retained by law or for security is quarantined and not used for training. Operational processing described above may continue for its separate purposes.
Completed-model limitation
Model training combines many examples and may not preserve an individual contribution in a separately retrievable form. We therefore cannot promise to remove a contribution already absorbed into a completed model. We will still take additional action when law requires it or lineage records make the relevant data technically separable, without claiming that an opt-out has no effect.
App scope and linked accounts
The preference applies to befall by default. It extends to another app only when the user expressly identifies the additional account and requests broader application. We do not link users across apps or create a new cross-app profile merely to propagate an opt-out. The request affects future training use and does not silently perform account or conversation deletion.
Document history
- 2026-07-27-draft.1 · 2026-07-27 — Initial PROJECT82 regional legal draft.
Last updated 2026-07-27 · content 4f29e865f3d4 · 대한민국 · 한국어 · 日本 · 日本語