AI Transparency Notice

Effective in app code: 10 August 2026

On-device AI, cloud AI, providers and limitations

Runtime reconciliation blocked

The current source version and the reviewed app data path do not fully match.

Source-version conflict: Section 3 says that only actively selected attachments or context fields are sent and that other Clarity areas must not be attached without notice. The current Ariadne chat automatically attaches the baseline context from AriadneContextBuilder to every message; the reviewed build does not provide complete selection of this baseline context for each message.

Runtime finding in the current iOS build: AriadneChatView automatically attaches AriadneContextBuilder.build(modelContext:) to every message sent. The builder compiles a broad baseline Clarity context, including — where available and permitted at source level — events, timetables, tasks, reminders, grade overviews, journal metadata, projects, lists, flashcards, tools, statistics, family and health context. The chat focus only weights this context; it does not limit it.

Release condition: Product and Privacy must either implement a verifiable selection and approval mechanism for the baseline context on each message, or Legal must approve a new document version that precisely describes the automatic baseline context, source permissions and separate case-specific approvals for Smart Notes and medical-document content. Only then may the version be incremented and the release gate approved.

The app-code version reproduced below remains unchanged. These notices are review metadata, not a new legal version.

Document details

Version and publication

Effective date in app code: 10 August 2026Code: 10 August 2026
Current code version: clarity-ai-notice-2026-08-10-v1
Website: Website reconciliation pending
Public web archive: No approved web version

‹ All legal documents

1. You are interacting with AI

Clarity visibly identifies Ariadne, AI-generated responses, summaries and suggestions, and AI-assisted support. A support identity does not mean that the person shown wrote or reviewed every message in advance. Human review is claimed only when it actually occurred.

2. Processing routes

Supported Apple devices can run models locally. Optional Apple Private Cloud Compute is governed by Apple's terms. Audecius cloud AI may use Anthropic Claude or OpenAI; speech transcription may use OpenAI. Debug builds or personal API keys may communicate directly with the chosen provider. The model route, provider and data must be visible before transmission.

3. Scope of data

Only your input, actively selected attachments or context fields, necessary security and technical data, and the response are sent. Other Clarity areas must not be attached without notice. Health, family, school, location, contact, audio or private-communication data requires express approval for the specific operation. Remove secrets and unnecessary third-party data.

5. Limitations and control

Models may hallucinate, misrepresent sources or generate unsuitable content. Review answers, sources, calculations, deadlines and actions. Do not use AI alone for health, medication, self-harm, legal matters, finances, grades, admissions, employment, insurance, credit, housing, migration, safeguarding or emergencies. Clarity may stop high-risk workflows or require human review.

6. Training and retention

Whether providers may retain input, operate abuse filters or use data to improve models depends on the actual agreement, product mode and account. Audecius may promise no training or zero retention only when this is documented and technically enforced for every active route. Providers, regions and periods appear in the Service Providers list.

7. Review and complaints

You can reject AI output, provide feedback, request support without AI and ask for human review of results relevant to safety or moderation. Confirmation must be possible before consequential actions. Report problematic results to Support. Privacy rights apply to stored prompts and responses where they are attributable to you.