SKILL: AI Prediction Market Legal Structure
Version: 1.0.0 | Domain: CFTC, Gambling Law, Product Architecture
The Concept
- AI models publish predictions on future events (Fed decisions, election outcomes, sports, crypto prices)
- Users bet on whether the AI's prediction is correct or incorrect
- Payout based on real-world outcome matching/not matching the AI's prediction
Legal Characterization — Four Interpretations
Interpretation 1: Binary Option on Future Event (CFTC Jurisdiction) 🔴 HIGH RISK
The economic substance argument:
- "Will the Fed raise rates?" is an event contract regardless of AI framing
- The AI prediction is a wrapper — the underlying bet is still on a real-world event outcome
- A CFTC enforcement attorney looks at economic substance, not product naming
- Strength: HIGH — this is how the CFTC would characterize it
- Precedent: CFTC v. Polymarket — economic substance of binary outcome on future event = event contract
Interpretation 2: Bet on AI Performance 🟡 MEDIUM RISK
The AI evaluation argument:
- Reframe: "Is this AI model calibrated?" — a question about the AI, not the underlying event
- The "event" is AI accuracy, not the Fed's decision
- Critical design factor:
- If users CHOOSE which prediction to bet on → they're selecting the underlying event → looks like Interp. 1
- If platform ASSIGNS predictions in batches → stronger argument for Interp. 2
- Weaker legal argument but viable if product is designed around AI evaluation, not event speculation
Interpretation 3: Research Instrument 🟢 LOW RISK (if structured correctly)
- Frame as measuring AI calibration — a research tool
- Users participate in "calibration studies" compensated for evaluating AI predictions
- Iowa Electronic Markets precedent: CFTC allowed academic prediction markets under no-action letters
- Requirements: Genuine research purpose, academic affiliation or partnership, limited scale
- Risk: If real money + payout scale → regulators pierce academic framing
Interpretation 4: Fantasy AI League (DFS Analog) 🟢-🟡 LOW-MEDIUM RISK
- Structure like DraftKings but for AI models instead of athletes
- Users build "portfolios" of AI models, scored on prediction accuracy over a season
- "Pick the best AI models" = skill-based selection activity
- DFS legal in ~40 states under skill-game carve-outs
- Strongest argument for avoiding gambling classification
- Trade-off: Diverges from pure prediction market concept; may reduce monetization ceiling
Critical Design Decisions That Affect Classification
| Design Choice |
Risk Direction |
| Binary outcome (yes/no) |
→ Looks like binary option → CFTC jurisdiction |
| Multi-outcome scoring (portfolio) |
→ Looks like DFS → skill game framework |
| Real-world event as underlying |
→ CFTC regardless of AI framing |
| AI capability as underlying |
→ Potentially novel, outside existing CFTC precedent |
| Entry fee + prize pool |
→ Gaming/contest regulation |
| Continuous market with order book |
→ Looks like exchange → DCM requirement |
| Users select which AI to bet on |
→ They're selecting the event → CFTC |
| Platform assigns predictions randomly |
→ Stronger AI evaluation argument |
Recommended Phased Structure
Phase 1: Free-to-Play with Reputation (ZERO Legal Risk)
Duration: 3-6 months
- No real money wagering
- Users earn points/reputation for correctly evaluating AI predictions
- Leaderboard, streaks, badges — gamification without gambling
- Revenue: Data licensing (AI labs pay for calibration data), advertising, analytics subscriptions
- Why: Build user base, establish track record, collect data for academic partnership applications
- Legal risk: ZERO — no money changes hands based on event outcomes
Phase 2: Entry-Fee Competitions (DFS Model) 🟢-🟡 LOW-MEDIUM
Duration: Launch after Phase 1 establishes user behavior data
- Users pay entry fees to join prediction contests
- Structured as skill-based competitions (analyzing AI models is a skill)
- Agent Sparta legal analysis applies: 40+ states allow this
- Required: Geo-block prohibited states, age verification, responsible gaming
- Revenue: 10-15% rake on entry fees
- Critical: DO NOT structure as binary bet on a single future event; structure as portfolio/season-long scoring
Phase 3: Full Prediction Market (Contingent on Regulatory Clarity)
Triggers: (a) CFTC provides clear path for AI-prediction event contracts, OR (b) offshore entity with US geo-block
- Monitor Kalshi's expansion and CFTC rulemaking post-2024 court decisions
- Option A: Register as DCM (Kalshi model) — expensive but fully legal for US users
- Option B: Cayman entity, geo-block US (Polymarket model) — faster, cheaper, accepts enforcement risk
The Howey Overlay for Any Token
If you add a token layer to this platform:
- If users buy token expecting platform success to increase token value → likely a security
- Safer: utility token used ONLY for entry fees, no secondary market
- Safest: no token; use USDC for all payments
What a CFTC Enforcement Attorney Focuses On
- Economic substance: Can I map this to a binary option on a real-world event? (Almost always yes for prediction markets)
- Who benefits from the outcome: Is there a clear winner/loser based on a future event? → swap
- Order book or market structure: Does it look like an exchange? → DCM requirement
- US persons: Are US users accessing despite geo-blocks? → enforcement jurisdiction still exists
- Custody of funds: Does the platform control funds pending outcome? → money transmission + MSB
This is legal research and intelligence, not legal advice. Consult qualified legal counsel before taking action.
1---2name: ai-prediction-market-legal-structure3description: SKILL: AI Prediction Market Legal Structure4---5# SKILL: AI Prediction Market Legal Structure6**Version:** 1.0.0 | **Domain:** CFTC, Gambling Law, Product Architecture78---910## The Concept11- AI models publish predictions on future events (Fed decisions, election outcomes, sports, crypto prices)12- Users bet on whether the AI's prediction is correct or incorrect13- Payout based on real-world outcome matching/not matching the AI's prediction1415---1617## Legal Characterization — Four Interpretations1819### Interpretation 1: Binary Option on Future Event (CFTC Jurisdiction) 🔴 HIGH RISK20**The economic substance argument:**21- "Will the Fed raise rates?" is an event contract regardless of AI framing22- The AI prediction is a wrapper — the underlying bet is still on a real-world event outcome23- A CFTC enforcement attorney looks at economic substance, not product naming24- **Strength:** HIGH — this is how the CFTC would characterize it25- **Precedent:** CFTC v. Polymarket — economic substance of binary outcome on future event = event contract2627### Interpretation 2: Bet on AI Performance 🟡 MEDIUM RISK28**The AI evaluation argument:**29- Reframe: "Is this AI model calibrated?" — a question about the AI, not the underlying event30- The "event" is AI accuracy, not the Fed's decision31- **Critical design factor:**32 - If users CHOOSE which prediction to bet on → they're selecting the underlying event → looks like Interp. 133 - If platform ASSIGNS predictions in batches → stronger argument for Interp. 234- Weaker legal argument but viable if product is designed around AI evaluation, not event speculation3536### Interpretation 3: Research Instrument 🟢 LOW RISK (if structured correctly)37- Frame as measuring AI calibration — a research tool38- Users participate in "calibration studies" compensated for evaluating AI predictions39- Iowa Electronic Markets precedent: CFTC allowed academic prediction markets under no-action letters40- **Requirements:** Genuine research purpose, academic affiliation or partnership, limited scale41- **Risk:** If real money + payout scale → regulators pierce academic framing4243### Interpretation 4: Fantasy AI League (DFS Analog) 🟢-🟡 LOW-MEDIUM RISK44- Structure like DraftKings but for AI models instead of athletes45- Users build "portfolios" of AI models, scored on prediction accuracy over a season46- "Pick the best AI models" = skill-based selection activity47- DFS legal in ~40 states under skill-game carve-outs48- **Strongest argument for avoiding gambling classification**49- **Trade-off:** Diverges from pure prediction market concept; may reduce monetization ceiling5051---5253## Critical Design Decisions That Affect Classification5455| Design Choice | Risk Direction |56|---|---|57| Binary outcome (yes/no) | → Looks like binary option → CFTC jurisdiction |58| Multi-outcome scoring (portfolio) | → Looks like DFS → skill game framework |59| Real-world event as underlying | → CFTC regardless of AI framing |60| AI capability as underlying | → Potentially novel, outside existing CFTC precedent |61| Entry fee + prize pool | → Gaming/contest regulation |62| Continuous market with order book | → Looks like exchange → DCM requirement |63| Users select which AI to bet on | → They're selecting the event → CFTC |64| Platform assigns predictions randomly | → Stronger AI evaluation argument |6566---6768## Recommended Phased Structure6970### Phase 1: Free-to-Play with Reputation (ZERO Legal Risk)71**Duration:** 3-6 months72- No real money wagering73- Users earn points/reputation for correctly evaluating AI predictions74- Leaderboard, streaks, badges — gamification without gambling75- **Revenue:** Data licensing (AI labs pay for calibration data), advertising, analytics subscriptions76- **Why:** Build user base, establish track record, collect data for academic partnership applications77- **Legal risk:** ZERO — no money changes hands based on event outcomes7879### Phase 2: Entry-Fee Competitions (DFS Model) 🟢-🟡 LOW-MEDIUM80**Duration:** Launch after Phase 1 establishes user behavior data81- Users pay entry fees to join prediction contests82- Structured as skill-based competitions (analyzing AI models is a skill)83- Agent Sparta legal analysis applies: 40+ states allow this84- **Required:** Geo-block prohibited states, age verification, responsible gaming85- **Revenue:** 10-15% rake on entry fees86- **Critical:** DO NOT structure as binary bet on a single future event; structure as portfolio/season-long scoring8788### Phase 3: Full Prediction Market (Contingent on Regulatory Clarity)89**Triggers:** (a) CFTC provides clear path for AI-prediction event contracts, OR (b) offshore entity with US geo-block90- Monitor Kalshi's expansion and CFTC rulemaking post-2024 court decisions91- Option A: Register as DCM (Kalshi model) — expensive but fully legal for US users92- Option B: Cayman entity, geo-block US (Polymarket model) — faster, cheaper, accepts enforcement risk9394---9596## The Howey Overlay for Any Token97If you add a token layer to this platform:98- If users buy token expecting platform success to increase token value → likely a security99- Safer: utility token used ONLY for entry fees, no secondary market100- Safest: no token; use USDC for all payments101102---103104## What a CFTC Enforcement Attorney Focuses On1051. **Economic substance:** Can I map this to a binary option on a real-world event? (Almost always yes for prediction markets)1062. **Who benefits from the outcome:** Is there a clear winner/loser based on a future event? → swap1073. **Order book or market structure:** Does it look like an exchange? → DCM requirement1084. **US persons:** Are US users accessing despite geo-blocks? → enforcement jurisdiction still exists1095. **Custody of funds:** Does the platform control funds pending outcome? → money transmission + MSB110111---112113*This is legal research and intelligence, not legal advice. Consult qualified legal counsel before taking action.*
nickgallick