Counsel review draft · version 2026-07-25

AI-Processing Disclosure

This draft explains the intended use and limits of artificial intelligence in the PAV entrepreneurship platform. External AI processing and applicant workflows remain disabled pending provider, privacy, security, and licensed-counsel approval.

1. Where AI may assist

• Interactive interview turns, clarification prompts, and stage detection.
• Evidence-grounded summaries, missing-information lists, consistency review, verification suggestions, venture-fit organization, and applicant-specific interview questions.
• Drafting applicant-safe Readiness and Fit Profiles after required human review and approval.
• Pilot calibration that compares human and AI observations without exposing internal calibration material to applicants.

2. Information AI may receive

Approved evidence inputs are limited to relevant interest-form fields, versioned application answers, approved workflow metadata, interview responses, applicant-confirmed summaries, and authorized human-review records. Evidence citations must identify stable source fields, question IDs, answer versions, or authorized records. Unsupported conclusions must be identified as insufficient evidence.

Uploaded files and extracted attachment text are excluded from external AI processing unless a later, separately approved contract and feature gate expressly authorizes them. Hidden chain-of-thought and private model reasoning are not requested or stored.

3. Human-controlled decisions

AI recommendations are advisory. AI cannot automatically accept, decline, reroute, release a profile, establish integrity concerns, or create a paid engagement. Kyle or another authorized human reviewer controls advancement, requests for information, holds, declines, paid-pilot decisions, assessment release, opportunity recommendations, and other consequential outcomes.

AI failure, timeout, or unavailable evidence cannot prevent authorized human review.

4. Internal AI-assistance likelihood estimate

PAV may create an internal estimate that a response appears to have received AI assistance. The estimate remains administrator-only and is excluded from applicant profiles, emails, exports, and applicant routes. It cannot automatically affect readiness, routing, acceptance, decline, integrity, or PAV-fit outcomes.

An estimate may be used only to generate evidence-linked, person-to-person clarification questions. Vague or polished writing is not treated as dishonesty, and no adverse outcome may rest on the estimate alone.

5. Fairness and prohibited inference

PAV does not authorize AI to infer merit from protected characteristics, age, disability, accent, writing polish, education prestige, family status, eye contact, or other irrelevant traits. AI must not diagnose mental-health conditions, personality disorders, honesty, or deception. Evidence labels remain qualitative and do not create one total founder score.

6. Providers, safeguards, and records

If approved, PAV plans to use a dedicated PAV OpenAI API project with server-side configurable models, bounded reasoning levels, timeouts, retries, and `store: false` requests. Provider terms, data-processing commitments, retention, regional processing, incident handling, and subprocessor disclosures require approval before real information is transmitted.

PAV records the prompt and review version, provider/model identifier, input-source manifest, evidence citations, status, confidence, errors, reveal history, superseding history, and administrator actions. Logs and monitoring must exclude prompts, transcripts, answers, tokens, credentials, message bodies, signed URLs, and uploaded-file content.

7. Applicant notice, review, and contact

Counsel must determine the final notice timing, consent or opt-out requirements, accessible non-AI alternatives, correction and contest rights, explanation standards, and jurisdiction-specific automated- decision disclosures. Questions or privacy requests may be sent to sales@payneagenticventures.com. This draft has no legal effective date.