Kalshi has built an internal AI agent, Harrison, powered by Anthropic's Claude model, to test prediction market contract wording.
According to Bloomberg, cofounder Luana Lopes Lara said the tool, built on Anthropic's Claude model, is used across the company's workforce of 150 people, with the markets team identified as its heaviest user outside engineering.
Kalshi lists markets on outcomes such as elections, sports results and award ceremonies, and wagers worth several million dollars often hinge on the precise wording of a contract or the evidence source used to settle it. When the company was founded, Lopes Lara and her cofounder recruited debate graduates from Yale University to scrutinise contract structures, a practice reflected in the fact that one of those graduates still works at the company. Harrison now performs a comparable function, reviewing draft contracts, flagging weaknesses, and recommending templates or amendments before a market goes live.
Lopes Lara said the AI effectively operates as an engineer within the markets team, continuously testing certifications to identify where a contract's wording could create ambiguity. Kalshi has built a library of more than 500 templates for possible market types, each assessed for how broadly it can be reused, how it withstands stress testing, and whether it meets user requirements. The company has previously had to resolve contracts where wording did not match real world outcomes, including a market on whether a Netflix executive would say 'Warner Bros' during a January 2026 earnings call, which Kalshi settled as 'no' after the executive used the phrase 'Warner Brothers'.
Listing, resolution, and trading volumes
Listing a new market typically involves two staff members, one populating the template and a second reviewing it, after which a contract faces a delay of one to two hours before going live, with a bounty offered to users who identify flaws. Resolving a market follows a similar pattern, as one team member inputs an outcome and a second adds an independent determination, with Harrison checking whether the two responses match its own suggested answer. More complex cases, such as a Supreme Court ruling, involve an additional layer of review that can include the company's chief regulatory officer.
Furthermore, Lopes Lara said the process has become more efficient, as the AI now proposes which market and template to use for each new suggestion, highlights potential issues and recommends new certifications or amendments where required. Beyond contract management, Harrison also aggregates news coverage, analyses competitor offerings and recommends where Kalshi should focus new listings or liquidity incentives.
Trading activity on the platform has grown alongside interest in sports events. Kalshi recorded nearly USD 18 billion in notional trading volume in May 2026, driven by demand linked to the World Cup and NBA Finals, according to data compiled by users on Dune Analytics. In the first week of the World Cup in June 2026, the exchange recorded a weekly volume of USD 5.1 billion.
