AI Sports Betting
What happens when an AI bets on sports with real money: the Mythos experiment, where a frontier model was given $10,000 and built its own trading bot for the 2026 World Cup, plus how AI betting systems are designed and what they can and can't do.
More guides in this topic
An autonomous AI trading agent that designed itself: we gave Claude Fable 5 a vague goal and $10k, and it architected its own prediction-market bot. Here's the four-layer pattern and the safety envelope you wrap around an AI-built system that touches real money.
Every AI World Cup prediction names a winner. The one we gave $10,000 doesn't. It hunts mispriced odds, backs underdogs, and abstains when there's no edge. Here's why.
Related guides
Complete 2026 developer guide for SX Bet. REST endpoints (no API key needed for reads), Centrifugo WebSocket for real-time order books, EIP-712 signed writes, parlay RFQ flow, MCP server for AI agents, rate limits, and reference implementations.
Peer-to-peer sports betting means bettors trade directly against each other, no bookmaker setting prices. Here's how the model works, why it produces better odds, and the platforms running it in 2026.
Betting exchange vs sportsbook compared honestly. Exchanges win on pricing, treatment of winners, and trading flexibility; sportsbooks win on market breadth, simple UX, and promos. Here's the decision framework.
A betting exchange is a marketplace where bettors trade against each other instead of betting against a bookmaker. Here's how the two-sided model works, why exchange odds tend to beat sportsbooks, and how SX Bet's binary-outcome model differs from Betfair.
An open-source aggregator that pulls live odds from SX Bet and Polymarket, adjusts for fees, and routes each bet to the bes odds. Built with Claude Code.
Sports betting MCP server tutorial: wire SX Bet's live odds API into Claude Code or Cursor using FastMCP. Full Python implementation, tool definitions, Claude Code config.
How to bet on the 2026 World Cup on an exchange: trade outright and match markets against other bettors instead of a bookmaker, read odds as implied probability, and understand why exchange prices beat sportsbook futures.
About AI Sports Betting
Predicting winners vs trading prices
Most "AI sports betting" coverage is a model naming who will win a match. That is the weakest use of an AI in betting: match outcomes are noisy, and a confident pick says nothing about whether the price on offer is any good.
The version that matters is an AI trading markets with money: estimating a fair price, comparing it with the market, and betting only when the two disagree by enough to be worth it. A bet on a longshot can be a good bet even though the longshot will probably lose.
The Mythos experiment
Ahead of the 2026 World Cup, Claude Fable 5 was given a $10,000 USDC wallet and one instruction: make money trading the tournament. It didn't place bets itself. It designed an autonomous bot, which named itself Mythos, that priced matches from prediction-market prices and exchange order flow, sized positions to survive a bad run, and traded on SX Bet with every position settled on-chain.
The guides in this topic cover what the bot did, how it was built, and why an AI trading system needs a betting exchange rather than a sportsbook: an open API, no account limits, and settlement anyone can verify.