AI Governance Policy
Core Commitment: We use AI to amplify human expertise. Every AI-generated output undergoes mandatory human review before reaching clients, regulators, or the public.
The Medvi Case Study (2024-2026)
AI-powered telehealth company built a $401 million business with just 2 employees using ChatGPT, Claude, Grok, Midjourney, Runway, and ElevenLabs. Result: FDA Warning Letter (Feb 2026, #721455) and FTC investigation due to:
- AI-generated fake doctors
- Hallucinated drug prices
- False testimonials
- 5,000+ fake doctor ads across Meta platforms
The "Medvi Test" — Mandatory 5 Questions
| # | Question | If YES → |
|---|---|---|
| 1 | Could this be mistaken for a real person's work? | Add attribution: "AI-generated, human-reviewed" |
| 2 | Could this affect a customer's financial or legal decision? | Mandatory human expert review |
| 3 | Could this be used as evidence in court or regulatory review? | Primary source documentation required; AI only for formatting |
| 4 | Would we be comfortable if a journalist exposed this as AI-generated? | Don't use AI |
| 5 | Does this require a disclaimer about accuracy? | Don't use AI — use human expert + Information Weight system |
Tiered AI Usage Framework
| Tier | Use Case | Human Checkpoint | Examples |
|---|---|---|---|
| Tier 1: Production | Client-facing, regulatory, financial | Mandatory expert review | Risk scores, compliance reports, regulatory submissions, pricing models |
| Tier 2: Internal | Internal productivity, research | Manager review | Meeting notes, email drafts, research aggregation, presentation drafts |
| Tier 3: Prohibited | Fake personas, testimonials, regulatory claims | NEVER | AI-generated experts, fake reviews, unverified claims, false testimonials |
Medvi's Failures vs. Our Safeguards
| Medvi Misuse | Consequence | Our Alternative |
|---|---|---|
| AI fake doctors (Midjourney + GPT) | Consumer deception, FTC Act violation | Real Experts Only — verifiable credentials |
| AI face-swapped before/after photos | Deceptive advertising, FDA misbranding | 3-Source Validation for every significant claim |
| Unchecked AI customer service | Hallucinated prices, false claims, class action | Human-in-the-Loop — AI drafts, humans approve |
| AI content with accuracy disclaimer | No accountability, brand destruction | Information Weight System (A-E grading, no disclaimers) |
| 5,000+ unmonitored fake doctor ads | FTC investigation, affiliate shutdown | Partner KYC/KYB + real-time monitoring |
| No compliance review layer | FDA Warning Letter 721455 | Compliance-by-Design from day one |
Key Systems in Place
- Information Weight System: Claims graded A-E (Verifiable → Speculative) — transparent confidence levels, no accuracy disclaimers
- Immutable Audit Trail: All decisions, scores, verifications stored on ICP blockchain — 7-year retention, $5/GB/year
- 3-Source Validation: Cross-verification across three independent credible sources required
Implementation Roadmap
| Phase | Timeline | Actions |
|---|---|---|
| Foundation | Months 1-3 | Define policy; establish human checkpoints; role-based AI access controls; create "AI Usage Log" |
| Operational | Months 4-6 | Deploy tiered framework; train staff on Medvi Test; dedicated compliance officer; automated monitoring for client-facing content |
| Scale | Months 7-12 | AI-assisted research (human fact-check mandatory); AI-powered test automation (human QA mandatory); AI-generated internal analytics (human interpretation mandatory); quarterly audits |
The Golden Rule
"If Medvi did it with AI, we do it with human oversight.
If Medvi didn't review it, we triple-review it."
Medvi vs. Truth Meter + Sentinel Nexus
| Dimension | Medvi | Truth Meter + Sentinel Nexus |
|---|---|---|
| Speed to Launch | 2 months (no compliance) | 6 months (compliance-first) |
| AI Usage | AI replaces humans | AI assists humans |
| Content Quality | "AI slop", hallucinations | Information Weight A-E grading |
| Compliance | Afterthought → FDA warning | Day-one architecture |
| Expert Claims | Fake AI doctors | Real verified experts only |
| Accuracy | Disclaimed (no accountability) | Guaranteed with audit trail |
| Oversight | 2 employees, no compliance | Dedicated compliance team |
| Result | $401M → regulatory disaster | Sustainable, regulator-approved |
Accountability & Contact
| Role | Contact |
|---|---|
| Compliance Officer | eddy@highperformanceadvisory.com |
| Technical Lead | dzmitry@arli.ai |