Kalshi and Polymarket move into AI prediction contracts

Bloomberg reported this week that Kalshi and Polymarket are pursuing event contracts tied to artificial intelligence outcomes — model releases, benchmark results, capability milestones. The move is deliberate and the timing is not accidental.

The prediction market industry is in the middle of a litigation cycle that has absorbed most of its legal budget and most of its public attention. Kalshi's federal preemption argument is still working through the courts. Polymarket is managing a CFTC investigation. Both operators know that the next durable revenue line cannot sit inside the same regulatory dispute that is currently consuming them. AI contracts are the exit door they are building while the other doors are on fire.

The structural case for AI event contracts is real. A model release date, a benchmark crossing, a capability evaluation — these are binary outcomes with verifiable resolution conditions and genuine information asymmetry between insiders and the public. That asymmetry is exactly what prediction markets price well. When one party knows something the market doesn't, the contract price moves. That movement is the signal. Hayek's point, applied to the one domain where knowledge is most aggressively concentrated and most aggressively protected.

But the concentration is the problem too. AI capability assessments are not like election results. There is no independent returning officer. The organizations that develop the models also run most of the benchmarks. A contract resolving on whether a model crosses a defined threshold on a defined test is only as clean as the test's independence from the people who built the model. At the moment, that independence is contested at best.

I don't think the consensus take on this — that AI contracts are the obvious next growth category — lands where the reporting assumes it will. The operators most exposed to AI contract mispricing are not the sophisticated quants who will arb the edges. They are the retail participants who will treat a model release date as a sports score and a benchmark result as a final whistle. When the resolution condition turns out to be more complicated than the contract implied, those are the traders who lose and then call their congressman. The industry has enough congressional attention already.

The more durable opportunity is narrower: contracts on external AI events where no single party controls resolution. Regulatory approvals, government adoption decisions, published academic evaluations from independent labs. Those resolve cleanly. They also happen less frequently and attract less volume. That tradeoff — cleaner markets, smaller books — is not the story either operator is trying to tell right now while fundraising and litigation are both active.

What Kalshi and Polymarket are actually selling is the category, not the specific contracts. The message to investors and regulators is that prediction markets belong inside the AI economy's information infrastructure. Whether the individual contracts that follow from that message will resolve cleanly is a question neither operator has fully answered in public.