Prediction markets face a significant challenge in accurately resolving contracts, especially in contentious situations like Venezuela’s presidential election. When the government declared Nicolás Maduro the winner amid allegations of fraud, the market had to decide whether to follow official results or a consensus of credible reporting. This highlights a broader issue: flawed resolution mechanisms can undermine trust, leading to reduced participation and skewed price signals. Several examples illustrate these problems. In the Ukraine case, contract outcomes were influenced by manipulated online maps. During the U.S. government shutdown, a delay in updating an official website led to incorrect payouts. A contract about Ukrainian President Zelensky’s attire flipped from “Yes” to “No” after initial resolution, raising concerns about conflicts of interest. These incidents show that when money is at stake, ambiguous situations can easily be exploited. A proposed solution is to use large language models (LLMs) as resolution judges. By locking the model and its prompt into the blockchain at contract creation, participants know exactly how outcomes will be determined. This setup could enhance resistance to manipulation, accuracy, transparency, and neutrality. While LLMs won’t eliminate all issues, they could significantly improve the reliability of prediction markets.