Prediction markets are designed to aggregate human judgment, but many persistent trading opportunities are being seized by systems far faster than any person. Brief mispricings — such as outcomes that don’t sum to 100% or short delays in market updates — create windows for arbitrage that automated systems exploit.
Rodrigo Coelho, CEO of Edge & Node, says bots already scan hundreds of markets per second, a role increasingly filled by AI-driven agents. “Capturing those opportunities requires monitoring thousands of markets and executing trades almost instantly, which is why they’re largely dominated by automated systems,” he told Cointelegraph. That makes prediction markets a natural target for AI systems designed to exploit fleeting pricing gaps automatically.
Arbitrage in prediction markets often hinges on latency. If markets take even a few seconds to reflect an event, execution bots can detect and bet on the correct outcome during that window. A recent academic study found frequent pricing inconsistencies on Polymarket, both within markets (probabilities not totaling 100%) and across related markets, estimating roughly $40 million extracted from these inefficiencies. Platforms like Polymarket have introduced measures such as taker fees, and outcomes aren’t always finalized immediately, which can reduce the reliability or profitability of some strategies.
Beyond simple arbitrage, AI agents could amplify market-manipulation risks. Large bets can move thin markets; more capable agents could replicate or scale that behavior. Coelho noted that a sizable stake can sway a market, pointing to a prior instance where a single large bet influenced a high-profile political market. Polymarket’s open interest peaked around the 2024 U.S. election and remains elevated, with politics, sports, and crypto among the leading topics.
Pranav Maheshwari of Edge & Node warned that as AI agents improve, guardrails will become more important. “Up until now, AI agents have medium capability and we give them a lot of permissions. With this medium capability, they have already started acting autonomously,” he said, adding that future agents with much higher capabilities will require restrictions on their permissions.
Trading automation itself is evolving from simple execution bots to AI-assisted systems that identify and act on opportunities in real time. Current tools are largely rule-based but are becoming more sophisticated. Archie Chaudhury, CEO of LayerLens, observed that most retail traders still use chatbots like ChatGPT or Gemini for research, while advanced users employ coding agents (e.g., Claude Code) or autonomous execution tools (e.g., OpenClaw) to build and run automated trading strategies.
As AI literacy grows, agents may democratize strategies once limited to institutions, though competition will remain fierce. Large institutions are already employing AI for trading, sometimes privately. Existing large language model architectures are well-suited to interpreting structured financial data, lowering the technical barrier for building automated trading systems that previously required specialized quantitative expertise.
These dynamics mirror broader crypto markets, where arbitrage increasingly depends on automation rather than human judgment. The competitive edge is shifting toward speed and integrated AI-driven execution. Those who adopt advanced automation and agents will likely outperform those who rely on manual or slower methods, underscoring the need for platform safeguards and regulatory attention as AI agents play a larger role in prediction-market activity.