# VenturePath AI Data Use Notice: Owner and Counsel Review Copy

**Version:** 0.1 · **Review date:** 25 July 2026

## Current early-access service

No live AI request is made from the current public service. Assisted-thinking controls use deterministic guidance prompts and do not send venture content to an AI provider.

## Production AI roles

AI may be used to research public sources, propose questions, challenge assumptions, summarize user-selected evidence, identify contradictions, prepare text from approved records, or help format an export. AI cannot conduct customer interviews, certify evidence, approve a gate, calculate authoritative financial/impact results, or replace qualified advice.

## Before every material AI task

The interface should show:

1. the task and why AI is useful;
2. the project records and external sources to be sent;
3. the provider/model class and material location/retention/training terms;
4. sensitive data excluded or redacted;
5. how the output will be labelled, stored, and reviewed; and
6. a reasonable non-AI route where feasible.

## Data rules

- Send the minimum approved excerpts, not an entire workspace by default.
- Pseudonymize customer research and remove contact details before AI use.
- Keep AI suggestions separate from approved facts and log prompt/template version, input references, provider/model, time, cost, safety flags, and user disposition.
- Never use confidential venture content for shared-model training or cross-customer benchmarking without a separate, specific, informed, revocable opt-in and approved governance.
- Do not make optional training consent a condition of service.
- Contractually verify provider retention, training, security, subprocessor, transfer, deletion, and incident terms before activation.
- Provide reporting, correction, rejection, and deletion paths.

## Human responsibility

Users must verify citations, numbers, assumptions, and applicability. High-stakes legal, tax, accounting, investment, employment, medical, safety, regulated, and impact claims require qualified human review. The product must expose uncertainty and unsupported claims rather than fill them with plausible prose.