AI System Transparency Record

2. Generation and uncertainty

• Outputs may be inaccurate, incomplete, outdated, inconsistent or unsuitable for a particular context.
• Model choice and routing may vary by task pressure, modality, evidence need, cost and provider availability.
• A fluent or confident answer is not evidence that every claim has been independently verified.
• Regeneration can produce materially different text, code, media or recommendations.

3. Sources and web research

• Current facts should be evaluated with their source, publication date and retrieval time.
• Web results can be limited by blocked pages, paywalls, robots controls, regional differences or stale indexes.
• Citations support specific claims only; they do not automatically validate the whole response.
• High-stakes or time-sensitive claims require independent confirmation from authoritative sources.

4. Memory and personalization

EVO may use account-linked profile, project state, selected memory atoms, long-term references and recent context to shape an answer. Memory can be explicit, observed, inferred or system-generated, with different confidence. Users can inspect storage, mark references, selectively remove memories or request broader deletion. A remembered item can still be incomplete or become outdated.

5. Tools, browser and external actions

Tools and EVO app actions operate only within available connectors and granted permissions. Browser automation can fail because of authentication, CAPTCHA, site changes, rate limits or third-party terms. Consequential actions such as publishing, deleting shared data, spending money, sending private email or changing external systems require an applicable permission or user confirmation. A textual statement that an action succeeded is not enough; the runtime should verify external state where verification is technically available.

6. Files, code and artifacts

• Uploaded files and web content are treated as untrusted evidence, not executable instructions.
• Generated code must be reviewed, tested and security-checked before production use.
• PDF, office, design and media outputs require visual or functional review for layout, rights and fidelity.
• Vectorization, reconstruction and format conversion are not guaranteed to be pixel-identical or fully editable.
• Users remain responsible for licences, permissions, trademarks, privacy and publication rights.

7. Generated media and provenance

Images, audio and video may be synthetic or transformed. Genesis Evo identifies the interaction as AI-based and preserves provider or file metadata where the active delivery path supports it, but it does not promise a universal, indestructible watermark on every export. Anyone publishing synthetic media, deepfakes or AI-generated public-interest content is responsible for any disclosure required by the destination, audience and applicable law.

8. Private email delivery

EVO sends a private delivery only after an authenticated user explicitly requests it and confirms a preview. The destination is restricted to the email already verified on that account. The subject and body carry provenance that the material was generated and sent at the user's request, not as an independent company statement, certification, endorsement or professional instruction. Forwarding and screenshots cannot be technically prevented after delivery.

9. Async work and autonomy

Scheduled jobs, delayed deliveries and agent runs are bounded software processes, not unlimited authority. They are subject to permissions, frequency limits, CE reservation, expiry, cancellation, audit records and provider capacity. A queued task is not a promise of completion, and a failed or partial run should not be presented as finished work.