Kalshi Builds AI Agent to Audit Contract Rules, Pushing Prediction Market Risk Control Toward Automation
Claire Weston
Prediction-market exchange Kalshi has built an internal AI agent called Harrison to audit contract language, assist settlement calls, and recommend new markets — a system becoming core infrastructure as the platform's monthly volume approaches $18 billion.
Why does contract wording need an AI fact-checker?
In prediction markets, the exact wording of a contract determines who wins millions of dollars in bets — a single word can flip the outcome.
Kalshi once ruled "no" on a market asking whether a Netflix executive said "Warner Bros." on an earnings call — because the executive actually said "Warner Brothers," not "Warner Bros."
This means → contract language is not legal pedantry; it is the settlement rulebook for real money. The more precise the wording, the fewer disputes.
Who is Harrison, and what does it actually do?
Harrison is an internal AI agent Kalshi built on Anthropic's Claude model. It reviews contract wording, aggregates news, analyzes competitors, and recommends which markets to list next or where to direct liquidity incentives.
Co-founder Luana Lopes Lara says Harrison "battle-tests" the entire contract certification process — simulating edge cases to surface potential loopholes before launch.
In plain terms = think of it as a QA inspector that never clocks out, poking holes in every contract before it goes live.
How many checkpoints does a new market pass through before going live?
Listing a new market typically requires two people: one fills in the contract template — rules, display settings, risk disclosures — and a second independently reviews it.
After that, there is a one-to-two-hour cooling period specifically for catching issues; the platform also pays bounties to users who spot loopholes.
This means → human dual-review + AI stress-testing + community bounties — Kalshi has built three layers of defense into contract listing.
How are markets settled — and what role does AI play?
Some markets settle automatically based on external data providers. When the AI detects a surge of news coverage on a topic, it alerts the team with a list of markets that may need a ruling.
Most settlements follow three steps: a team member enters a result → a second person independently records their own call → AI checks whether the two match and compares both against its own suggested answer.
For complex markets like Supreme Court rulings, additional review layers are added — sometimes involving Kalshi's chief regulatory officer.
How big is the scale this system supports?
Sports betting demand — driven by events like the World Cup and NBA Finals — pushed Kalshi to a record in May 2026, with notional volume approaching $18 billion for the month.
In the first week of the World Cup this month, Kalshi's weekly volume hit $5.1 billion, a new weekly record.
This reflects a prediction market moving from niche experiment to mainstream financial infrastructure — and contract quality control is the prerequisite for scaling.
Content is for reference only, not financial advice.