AI character privacy.Follow the conversation data.

Public characters are discoverable; unlisted characters require a link; private characters are available only to their creator.

The creator-access FAQ says other character creators cannot read your conversations. That statement does not establish staff access or provider processing.

Open question: Provider-access conditions were not established from the help articles in this review.

The training FAQ describes using content for model improvement. Its EEA/UK opt-out excludes new content from generative-model training; other improvement uses can continue.

Open question: The newer July summary refers more broadly to opt-out. Availability outside EEA/UK and treatment of earlier content need checking in the full policy and account.

A retention timetable was not established from the reviewed help articles.

Open question: The full policy did not return readable text in the research tool. No retention duration is inferred from Delete Account.

The help article documents permanent account deletion through account-data settings.

Open question: Removal from backups, previously shared content and training datasets was not established by this article.

Account settings document Export data on web and mobile.

Open question: The delivered archive and its contents were not inspected; account export is not proof of a portable character card.

Character-sharing and discovery controls are outside this policy review.

Open question: Inspect the sharing control separately from conversation processing.

The policy says chats are encrypted at rest and in transit, but not end-to-end. Servers and automated safety systems process content; authorized human review has stated exceptions.

Open question: Atlas has not audited access controls. The policy lists requested review and legal or terms-enforcement conditions.

A clear conversation-training commitment or opt-out was not established in the reviewed policy.

Open question: General service-improvement language and adapting replies in a chat do not establish whether conversations train shared models.

The policy retains chats and generated media until the AI or account is deleted, with broader legal, dispute and anonymous-data exceptions.

Open question: A complete timetable for every copy and exception was not established.

The FAQ schedules the final Kin for 24-hour deletion and retains logistical data. Individual message deletion is currently unavailable in that FAQ.

Open question: Deleting a Kin is not the same action as removing one message or ending every billing relationship.

The policy documents a self-serve personal-data export every 30 days under Profile → Preferences.

Public character-discovery and sharing controls were not examined in this review.

Open question: A conversation-privacy statement does not establish all character-sharing settings.

The policy collects chat and customization content and permits sharing for described business, service and legal purposes.

Open question: The review does not establish every staff-access condition or encryption architecture.

Nomipedia describes aggregate learning from conversations and feedback after removing identifying information.

Open question: This is a product-wiki explanation. A specific opt-out and treatment of all earlier content were not established.

The policy describes personal-data deletion about 28 days after confirmation, with exceptions including support correspondence archives.

Open question: No universal immediate-deletion promise is inferred.

Account Settings is the described deletion route; the policy preserves stated archival and legal exceptions.

Open question: Archive exceptions and copied content require separate consideration.

The policy makes access and transfer rights depend on applicable law.

Open question: A self-serve conversation export or its file contents were not established.

A private-character publishing scope was not established from the privacy notice.

Open question: Inspect character-sharing controls separately; this review covers data processing.

The notice permits moderation review and says third-party model providers or hosters may receive chat content.

Open question: The particular providers and account-level handling were not inspected.

The notice describes AI and moderation-model training, including possible human review of de-identified or anonymized interactions.

Account data is generally kept three years after last activity, with a stated possible shorter period; deletion requests and financial records have separate treatment.

Open question: This is not a three-year rule for every data category.

The notice describes account-data deletion requests; legal obligations and claims can limit erasure.

Open question: Closing an account alone is not described as immediate erasure of every record.

Its GDPR section describes conditional machine-readable portability through a data request.

Open question: No self-serve chat archive or universally available export format was established.

The multi-user setup separates user data, but filesystem access can reveal it. A user password is not a hosted-platform privacy guarantee.

Open question: Check who administers the instance and who can access its files.

Documentation says server data, including chats, is stored in plain text. The interface can connect to local or cloud models.

Open question: A cloud model, media engine or extension adds its own data handling. Local interface installation alone does not keep inference local.

Model-training use depends on the connected backend and other services.

Open question: Review the actual model provider; the interface documentation does not establish a cloud backend’s opt-out.

The instance keeps user files; the reviewed docs do not provide a universal expiry policy.

Open question: Check instance backups and each connected service’s retention independently.

A character-delete control is documented.

Open question: Erasure of chat files, backups, shared cards and connected-provider data must be considered separately.

Character export and a user-data folder backup archive are documented as separate controls.

Character-discovery and private-sharing controls were not established from this notice.

Open question: Check character visibility separately from the policy’s conversation-processing statements.

The notice allows access by operational staff and service providers, alongside stated legal and business exceptions.

Open question: Its non-sharing statement has exceptions; it is not a promise that only the user can access messages.

The notice lists text and voice messages as used to improve replies and develop AI.

Open question: The exact shared-model training scope and a conversation-training opt-out were not established.

Data is retained as needed for the service, with legal and legitimate-interest exceptions.

Open question: A fixed expiry period for all conversations and backups was not established.

Profile deletion or a support request can remove data, subject to retention exceptions. Previously disclosed data and caches may remain.

Open question: No immediate erasure of every copy or completed deletion was verified.

The EU section describes conditional machine-readable personal-data portability through support and identity verification.

Open question: This does not establish a self-serve chat archive, character-card export or availability for every user.

The profile guide ties shared character links to public visibility. The policy separately describes public account profiles and submissions.

Open question: A character, an account profile and a conversation have different audiences; the complete private-character access scope was not established.

The policy collects chats and uploads, allows authorized business access and shares data with service providers and stated other recipients.

Open question: Encryption in transit is not an end-to-end encryption commitment.

A specific conversation-training commitment or opt-out was not established in the policy.

Open question: General service-improvement language does not establish whether chats train shared models.

Data is kept as needed for stated purposes, including legal, accounting and reporting requirements.

Open question: No fixed conversation or backup expiry was established; policy date labels also need clarification.

The help guide separates Clear Chat from Delete Character. The policy describes account deletion and conditional erasure requests.

Open question: Removing a visible chat or character does not establish removal of all retained provider records.

The policy describes personal-data access and portability rights that depend on applicable law.

Open question: A self-serve conversation archive or portable character-card export was not established.

Technical docs describe public, link-accessible unlisted and creator-only private characters. Definition visibility is separate; hidden characters remain visible to moderators.

Open question: The help article lists only private and public options. Current unlisted availability needs checking in the creator.

The policy permits sharing with service providers and other stated recipients. Visibility docs address characters, not staff access to conversations.

Open question: Detailed chat-access conditions and encryption architecture were not established.

A specific conversation-training commitment or opt-out was not established in the reviewed policy.

Open question: Service-improvement language and character memory do not answer the shared-model training question.

The policy retains personal data as needed for its purposes, legal obligations, disputes and enforcement.

Open question: No fixed chat or backup timetable was established.

The help article describes permanent removal of chats, characters, personas and memories. The policy preserves stated legal-retention exceptions.

Open question: The help article says website subscriptions cancel with the account; externally managed subscriptions require separate cancellation. Erasure outcomes were not tested.

The policy offers a support route for personal-data access and removal requests.

Open question: A machine-readable delivery format or self-serve chat archive was not established.