Legal 13

Ginie AI and Automated Research Terms

Effective 18 November 2026 · Last updated 18 November 2026 · Version 2026.11.18

Contents (14 sections)
  1. 13.1 Scope
  2. 13.2 Accuracy and disclosure
  3. 13.3 No autonomous trading
  4. 13.4 Inputs and privacy
  5. 13.5 Tool and model controls
  6. 13.6 Member conduct
  7. 13.7 Saved research and updates
  8. 13.8 Escalation
  9. 13.9 Model limitations and truthful sourcing
  10. 13.10 Deterministic calculations and reproducibility
  11. 13.11 Privacy and sharing boundaries
  12. 13.12 No autonomous transactions
  13. 13.13 Saved research and human control
  14. 13.14 Reporting errors and escalation

13.1 Scope

Dear user, Ginie is an AI-assisted research feature that may help summarise authorised information, explain deterministic calculations, compare research scenarios, identify missing evidence and challenge user hypotheses. Ginie may assemble instrument-specific analysis using available inputs. These features are support tools, not an autonomous investment decision-maker or a guarantee of correct financial conclusions.

13.2 Accuracy and disclosure

Generative models can omit material facts, produce incorrect references, misread accounting context, misunderstand a prompt or provide inconsistent answers. Mathematical outputs should be generated by tested deterministic tools rather than by mental arithmetic from the model. Users should verify material figures against source documents and the disclosed calculation method. When evidence is missing or rights do not permit access, Ginie should explicitly decline or qualify its answer.

13.3 No autonomous trading

Ginie does not place trades, open brokerage accounts, transfer money, accept deposits or instruct external brokers on your behalf as part of the INRGIFT research service. Suggested research actions, updates to saved research and saved hypotheses should require appropriate user confirmation and authorisation. AI-generated material does not change the legal nature of any regulated service; applicable authorisation and suitability requirements remain relevant.

13.4 Inputs and privacy

Prompts may include an instrument, model assumptions, research questions and selected private notes. We use them to process the requested task and maintain the conversation where the feature is active. The deployed inference location, processors, retention, model identity and any cross-border transfer must match the Privacy Policy and technical reality. Private member research should not be retrieved for another member.

13.5 Tool and model controls

The application may constrain model outputs with source checking, calculation tools, version controls, timeouts, request quotas and safety validation. These measures reduce some risks but do not guarantee that every response is complete or correct.

13.6 Member conduct

Do not ask Ginie to steal confidential information, impersonate users, bypass paid entitlements, generate fabricated company documents, promote market manipulation or process personal data you are not entitled to share. We may enforce limits to protect lawful users, infrastructure capacity, licensed data and the integrity of research results.

13.7 Saved research and updates

A Ginie answer may be temporary until explicitly saved; saved research can preserve your own conclusions and selected evidence. Changes in market data may prompt review notices but should not silently rewrite your authored investment thesis. Source revisions and calculation-version changes can cause a previously saved result to differ from newly calculated results.

13.8 Escalation

If Ginie produces a material error, misleading citation or private-data disclosure, report it to support@inrgift.com with a minimal description, affected feature and date. Do not include unnecessary confidential data. We may investigate and limit an unsafe capability while a defect is corrected.

13.9 Model limitations and truthful sourcing

Generative systems may misunderstand a question, mix periods, invent facts, overlook contradictory evidence or produce flawed reasoning. Citations should correspond to actually retrieved, authorised source material, and answers should identify significant uncertainty or data gaps. If a claim cannot be verified, Ginie should say so rather than fabricate a source or confidence score.

13.10 Deterministic calculations and reproducibility

Ratios, returns, currency conversion, valuation models and financial aggregates must be computed using tested deterministic functions and labelled inputs. The explanation should identify currency, period, convention, source and rounding where material. A language model's generated number is not a validated financial calculation simply because it appears in a formatted answer.

13.11 Privacy and sharing boundaries

Prompts can contain user-authored notes and sensitive research intentions. The product must limit AI retrieval to the authorised account and explicitly distinguish personal history, shared research and general public data. A future model-provider change, external transfer or training use needs appropriate privacy notices and legal review. No claim of self-hosting should appear until the production runtime is confirmed.

13.12 No autonomous transactions

The AI assistant may suggest a research action, draft a comparison or ask for confirmation to save notes, but must not represent itself as placing securities trades, opening brokerage accounts, moving client funds or issuing binding investment instructions within INRGIFT research services. Any future agentic action with legal or financial consequences requires explicit permission, audit logging and a separate regulatory assessment.

13.13 Saved research and human control

The member controls the thesis and authorised changes to saved research. AI may identify events or offer a proposed update, but must not silently overwrite saved research when source data changes. Apply version checks, explain conflicting edits and keep a record of accepted changes in a way that does not expose private research to other users.

13.14 Reporting errors and escalation

Users need a way to report incorrect calculations, unsupported citations, privacy incidents or harmful outputs. Reports should capture the response and relevant context only with appropriate access controls, then route to the correct QA or security owner. The presence of human review or automated guardrails does not guarantee perfect answers.