AI bots exploit micro-arbitrage in short-term Polymarket prediction markets

Automated and AI-driven trading bots are increasingly capturing micro-arbitrage opportunities in ultra-short-term crypto prediction markets (notably Polymarket). Early reports described high-frequency bots exploiting mispriced 15-minute up-down markets and using temporal arbitrage, front-running thin order books, and buying both sides when combined Yes/No prices dipped below $1 to lock near-risk-free gains. A later, more detailed report documented a fully automated bot executing 8,894 trades on five-minute bitcoin (BTC) and ether (ETH) prediction contracts and netting nearly $150,000 by buying both sides when combined prices briefly summed to < $1, yielding ~1.5%–3% per round-trip (about $16.80 per trade). These inefficiencies exist for milliseconds and require low-latency infrastructure; order-book depth (~$5k–$15k per side) caps profitable trade size in the low four figures, limiting scalability for large capital. More advanced strategies compare implied probabilities from options and derivatives with prediction-market prices; AI agents automate strategy discovery, parameter tuning and multi-market monitoring. Reported win rates for some automated strategies exceed 85–98%, while human traders typically underperform due to late entries, oversized bets and weaker risk controls. As more algorithmic traders pursue these edges, spreads are expected to tighten and latency will become decisive, shifting prediction markets toward derivative-like pricing and reducing independent crowd-sourced signal value. For traders: opportunities remain for disciplined, small-size, low-latency micro-arbitrage; focus on strict position sizing, fast execution, monitoring of order-book depth and cross-market implied probabilities. Expect increased competition, narrower edges, and liquidity constraints that limit scalable deployment.