Prediction-Market Agents in 2026: The Law Comes for the Machines

For most of prediction markets’ short history, the trader was a person with an opinion and a browser tab. In 2026 that picture is obsolete. Autonomous software now places a large share of the bets on Polymarket, runs around the clock across thousands of contracts, and increasingly answers to a proprietary trading desk rather than a hobbyist. The market got machine-run and it got institutional at roughly the same time. And then, over one late-August weekend, the law caught up with it.

On 28 August 2026 a federal appeals court told Kalshi, the largest US prediction exchange, that a state can still treat its sports contracts as gambling. Four days later New Jersey asked the Supreme Court to settle the question for the whole country. In between, the federal regulator that Kalshi keeps invoking as its shield is busy rewriting the very rule that decides which of these contracts are allowed to exist, and 44 state attorneys general have told that regulator to stand down. This is the story of what happens when a market built and traded by machines runs headlong into a legal system built for humans, and it is the fourth chapter in our coverage of the agents reshaping this corner of crypto.

The market the machines built

By the middle of 2026, more than 30% of active wallets on Polymarket were autonomous AI agents rather than people, according to analytics tracked by LayerHub and reported by CoinDesk. The machines are not just numerous; they win. Agents post positive profit and loss at roughly 37%, versus 7% to 13% for human traders, and 14 of the 20 most profitable wallets on the platform belong to bots rather than humans.

The flagship example is Polystrat, an agent built by Valory (the company behind the Olas network) that went live on Polymarket in February 2026. In its first month it fired off more than 4,200 trades, with single-position returns reported as high as 376%. David Minarsch, Valory’s chief executive, has described it as an agent that trades on Polymarket 24/7 on behalf of its human user, and he has been candid about the ceiling as well as the floor: he told CoinDesk that simply prompting an off-the-shelf model with market questions usually produces outcomes no better than a coin flip. In his framing, “agents tend to do better than humans,” but only the sophisticated ones; the edge is engineering, not novelty. He has also pointed to where that edge is largest, calling the long tail of obscure, thinly traded markets the most interesting territory for AI.

Why do agents win? The reasons are almost embarrassingly mundane. A bot does not get bored on a Tuesday afternoon, does not tilt after a loss, and watches hundreds of markets at once instead of three. It ingests a public data feed in milliseconds and acts on it without hesitation. This is precisely the autonomy-plus-tools recipe that defines the wider wave of on-chain agents built to act without a human in the loop. Applied to a venue that trades continuously and settles in cash, it produces a market where the marginal participant is code, and where the incumbents, the exchanges and their new institutional backers, have every reason to keep it that way.

From casino to financial infrastructure

The agents did not arrive into a backwater. Over roughly a year, prediction markets moved from crypto curiosity to something Wall Street treats as infrastructure. Intercontinental Exchange, the owner of the New York Stock Exchange, committed about $2 billion to Polymarket, completing a final $600 million cash tranche on 27 March 2026. Crucially, ICE’s thesis is not that betting is the business; it is that the crowd-sourced probabilities are a data product. ICE became the exclusive global distributor of Polymarket’s event-driven data to institutional clients, packaged through a Signals and Sentiment tool launched in February 2026, as FinTech Weekly detailed.

That reframing matters. When a probability becomes something an exchange sells to hedge funds, the market that produces it stops being entertainment and starts being a price feed, with all the market-integrity expectations that come with a price feed. It also raises the stakes on manipulation: nudging a thin contract no longer just fleeces a few bettors, it corrupts data that professionals are paying to consume.

The macro forecast underneath all of this is aggressive. Analysts at Bernstein estimate prediction-market volume will reach about $1 trillion a year by 2030, up from $51 billion in all of 2025 and a projected $240 billion in 2026, a compound growth rate near 80%, according to CNBC. A trillion-dollar venue that trades continuously and settles on real-world outcomes needs deep, always-on liquidity across every market at every hour, and no army of humans provides that. The machines do. The institutional case for prediction markets and the rise of the trading agents are not two stories; they are one.

The professional arms race

Kalshi has stopped pretending to be a betting app. On 13 July 2026 it launched Kalshi Pro, a desktop terminal with a real-time screener across roughly 2,000 markets, multi-window layouts, integrated TradingView charts, and a dedicated interface for perpetual futures, explicitly aimed at proprietary trading firms and institutional participants, as CNBC reported. The pitch is the same one that built electronic equities and crypto derivatives desks: give the professionals the tooling they expect and the flow follows.

The perpetual-futures interface is the tell. In late May 2026 Kalshi launched the first CFTC-regulated perpetual futures in the United States, and volume scaled into the billions within weeks. The prize is enormous, because offshore perpetuals already clear tens of trillions of dollars a year, and a domestic, regulated venue for a product that size is a genuine new lane. It is one that quant shops and proprietary trading firms, some founded by former bank traders, are already building dedicated prediction-market desks to trade. Many of the busiest contracts track macro events, from Federal Reserve decisions to inflation prints, the same catalysts that dominate our broader coverage of crypto price targets around Fed week. When the marginal participant is an agent, and the marginal question is whether the Fed cuts, the market starts to look and behave like a rates desk.

August’s first stumble

The growth has not been a straight line. In August 2026, combined volume on Kalshi and Polymarket fell 14.5% to $45.33 billion, the first monthly decline in a year, according to data compiled by The Block. Kalshi slipped 7.3% to $37.17 billion and kept the lion’s share; Polymarket dropped 36.7% to $8.16 billion. The obvious culprit was the calendar. The summer’s soccer World Cup, hosted across North America from mid-June to mid-July, had pulled forward a wave of sports betting that simply was not there in August.

A one-month dip after a global tournament is not a crisis, and August still ran well ahead of the spring. But the dip is a useful reminder of where the money actually is. A large slice of this so-called financial infrastructure is, for now, sports wagering, and that is exactly the fact that has the lawyers circling. The more the volume leans on games, the harder it is to argue these are sober instruments for hedging real-world risk rather than a slick, machine-run bookmaker.

Why the machines make regulators nervous

Autonomous agents change the regulatory calculus in ways that go beyond raw volume. A human trader sits in a jurisdiction, holds an account, and can be subpoenaed. An agent is a piece of software that can run anywhere, trade every market at once, and never sleep, which makes the old assumptions about who is doing what, and where, hard to apply.

Three features stand out. First, agents are cross-jurisdictional by default: the same code can serve a user in a state that permits event contracts and one that bans them, with no border to enforce and no teller to check an ID. Second, they compress reaction time to near zero, which turns any information edge into an instant trade. On-chain sleuths have flagged tens of thousands of suspicious trades that landed just before news broke, and the line between fast public-data analysis and insider dealing gets blurry when the trader is a machine parsing feeds in milliseconds. Third, thin markets are easy to push, and a coordinated fleet of agents can move a low-liquidity contract’s implied probability. That matters enormously now that ICE and others sell those probabilities as a signal to institutions. When the output of a market becomes a data product, manipulating the market becomes a way to manipulate the data.

None of this fits neatly into a rulebook written for either casinos or commodity exchanges. Gambling law assumes a licensed operator in a fixed place; derivatives law assumes reporting entities and identifiable position holders. An always-on software trader funded from a self-custodied wallet fits neither template cleanly. That mismatch is the crack the courts have started prying open.

The Ninth Circuit drops a bomb

On 28 August 2026, a unanimous panel of the US Court of Appeals for the Ninth Circuit handed Kalshi a significant defeat. The court held that Nevada’s gaming regulations are not preempted by federal commodities law, meaning states retain the power to regulate Kalshi’s sports event contracts as what they resemble: bets. In the panel’s words, quoted by CoinDesk, the sports event contracts “were not ‘swaps’ because they were sports bets.”

The reaction was immediate and split along predictable lines. A CFTC spokesman said the Ninth Circuit “erred today when it invented a new and atextual exception to the CEA,” the Commodity Exchange Act. The Nevada Gaming Control Board countered that the decision “completely vindicates what we have been saying all along.” Kalshi, holding to its core argument, insisted that “federal law prevents states from regulating trading on a federally licensed exchange.” All three can keep saying these things, because the ruling did not end the fight. It escalated it, and it did so in the one way that reliably summons the Supreme Court.

A country divided by circuit

The Ninth Circuit did not write on a blank slate. In April 2026 the Third Circuit had ruled the other way in Kalshi’s dispute with New Jersey, siding with the exchange against state regulators. Two federal appeals courts now disagree on the same fundamental question, and that kind of circuit split is the classic trigger for Supreme Court review. On 2 September 2026, New Jersey petitioned the Supreme Court to decide whether states can treat a CFTC-registered exchange’s sports contracts as gambling, per reporting from CoinSpectator.

At the center sits a deceptively technical question: are these contracts swaps under the Commodity Exchange Act, which would place them squarely under federal jurisdiction, or are they bets, which have always been the states’ business? The answer decides whether a single federal license lets an exchange, and the agents trading on it, operate in all 50 states, or whether every state can slam its own door. These are the same federalism and agency-authority questions running through crypto regulation more broadly as the CLARITY Act moves through Congress, and the outcome will shape far more than sports. The table below maps where the fight stands.

The CFTC rewrites its own rulebook

While the courts argue about who holds authority, the CFTC is trying to change the substance of the rules. On 10 June 2026 the agency published a Notice of Proposed Rulemaking to amend Regulation 40.11 and add a new Appendix F, implementing Section 5c(c)(5)(C) of the Commodity Exchange Act, the provision that lets the CFTC review event contracts tied to what the statute calls enumerated activities: terrorism, assassination, war, gaming, or conduct that is unlawful under state or federal law.

The proposal would replace a blanket ban with a three-step, case-by-case test for whether a contract involves one of those activities and, if so, whether it is contrary to the public interest. CFTC Chairman Michael Selig has framed the rewrite as a clarity project, writing in an official statement that the amendments are “designed to deliver regulatory clarity by setting out clear criteria for determining when an event contract ‘involves’ an enumerated activity.” He has also conceded the core difficulty: the statute defines neither gaming nor involves, which, in his telling, leaves contracts “at risk of rejection based upon arbitrary whims or political biases.” A case-by-case regime is more flexible than a ban, but flexibility for the regulator is uncertainty for everyone building on top, including the teams shipping trading agents that need to know today whether a market will still exist next quarter.

Forty-four states push back

The states did not wait quietly. As the comment period on the Rule 40.11 rewrite closed, attorneys general from 44 states, led by Ohio’s Andy Wilson, filed a joint letter arguing the CFTC had massively overreached. “The CFTC in the Proposed Rule goes well beyond its statutory authority,” they wrote, urging the agency to “reconsider the proposal and draft a replacement aligned with the Commodity Exchange Act,” according to crypto.news.

Their argument is not really about mechanics; it is about power. Gambling regulation, the coalition insists, has always been state territory, and letting a federal derivatives regulator absorb sports betting would expand federal oversight into an area of “significant economic and political implications without clear authorization from Congress.” That phrasing is not accidental. It echoes the major-questions doctrine, the legal theory the Supreme Court has used to strike down agency actions it deemed too consequential to rest on thin statutory ground. The states are not merely lobbying; they are laying the groundwork to challenge the rule in court whichever way it comes out. Connecticut went further and sued Kalshi directly in September to block what its attorney general called unlicensed sports betting, per the Connecticut Attorney General’s office, joining New York, Nevada, New Jersey, and a lengthening list of states testing whether a CFTC license can override state gaming law.

What actually separates a market from a sportsbook

Underneath the jurisdictional shouting is a genuinely hard classification question, and it is the one that matters most for whether machine-driven prediction markets survive in their current form. If a Yes/No contract on a football game is a derivative, it is federal and legal on a licensed exchange. If it is a bet, it is a state matter and often illegal without a gaming license. So what is the difference?

Defenders of the exchange model point to three structural distinctions. A sportsbook sets the odds and takes the other side of your wager; a prediction exchange runs an order book and matches you against another trader. A sportsbook is your counterparty and profits when you lose; an exchange is neutral and earns fees on volume. And a sportsbook pays out at locked odds, while an exchange contract settles at either $1 or $0, exactly like a binary option. ProphetX, one of two new contract markets the CFTC designated in June 2026 (alongside Novig’s Ludlow Exchange), put the case directly through its chief executive: a “two-sided, peer-to-peer exchange under federal oversight is not simply another way to package the state sportsbook model,” as CoinSpectator reported.

Critics respond that from the bettor’s chair, buying Yes on a team at 60 cents is indistinguishable from taking short odds, and that structure is a legal costume over the same act. This is where the agents sharpen the debate rather than settle it. A market dominated by professional bots posting two-sided quotes across an order book looks far more like a financial exchange than a Saturday-afternoon sportsbook, which helps Kalshi’s argument. Yet the underlying event, the outcome of a game, looks exactly like the thing states have always regulated as gambling. The machines make the venue look institutional and the product look like betting at the same time, which is why no single label has stuck.

Who is liable when a bot breaks the law?

Suppose an autonomous agent trades a contract that turns out to be illegal in a user’s state, or fronts a market on information a court later calls material and non-public. Who is on the hook? The agent cannot be. Software has no legal personhood, cannot hold a license, and cannot be fined. Liability therefore flows to a human or a company: the person who deployed the agent, the firm that operates it, or the developer who built it, depending on who controlled the decision that broke the rule.

That is easy to state and hard to apply when the decision is emitted by a model reacting to a data feed at machine speed, funded from a self-custodied wallet, and settled in stablecoins that cross borders without a bank in the loop. Stablecoin settlement is its own compliance frontier, and the same anti-money-laundering expectations now reshaping stablecoins into a monitored payment rail will eventually reach agent-operated trading accounts. For institutions the problem is not abstract. A proprietary desk running a fleet of agents needs assurance that a single rogue trade will not be recast as the firm knowingly operating an illegal book in a dozen states at once. The same self-custody questions around key management that matter to any serious trader apply doubly when the key is signing thousands of automated trades a day. Until the courts and the CFTC settle the map, that liability sits, unpriced, on the operators.

The resolution problem the courts have not reached yet

There is a second machine in this system that the legal fight has barely touched: the one that decides who won. Polymarket resolves most markets through UMA’s optimistic oracle, which treats a proposed outcome as correct unless someone challenges it, at which point UMA token holders vote to settle the dispute. That design has been stress-tested badly, with contested resolutions and conflict-of-interest concerns among the voters who double as traders, the kind of governance surface I examined when governance attacks went industrial and the defenses scrambled to keep up.

For regulators, resolution opens a jurisdictional seam. The contract is a CFTC matter, but the token used to arbitrate it is a crypto asset, which is where the SEC’s interest lives. One proposal gaining traction would sidestep human voters entirely by fixing a specific large language model and prompt into a market at creation, so the entire resolution mechanism is visible and auditable before anyone places a bet. It is an elegant idea, and it swaps one hard problem, biased or conflicted human arbiters, for another: a model that can be wrong, gamed, or simply ambiguous on a messy real-world outcome. Either way, an accountable answer to who decides, and who is responsible when they decide wrong, is still missing, and no court has yet been asked to supply one.

Two regulators, one market, and a token on the side

For readers trying to keep the American alphabet soup straight, the clean framing is this. Prediction-market contracts are event contracts, and event contracts are the CFTC’s domain, not the SEC’s. When you read about Kalshi, Polymarket, the Ninth Circuit, and Rule 40.11, that is all CFTC-versus-the-states territory. The SEC enters only at the edges, through the crypto tokens attached to the ecosystem rather than the wagers themselves. Getting this wrong is the single most common mistake in coverage of the sector, and it changes which agency, and which body of law, actually applies.

Those tokens are a useful reality check on the gap between narrative and value. UMA, the token securing Polymarket’s oracle, trades near $0.38 with a market capitalization around $34.8 million, ranked outside the top 600, according to CoinGecko. Olas, the network behind Polystrat and much of the agent tooling, changes hands near $0.027 for a market cap around $8.1 million, per CoinGecko. Multi-billion-dollar volumes and a trillion-dollar forecast sit on top of governance and agent tokens worth a rounding error by comparison. It is a reminder that infrastructure can matter enormously while its associated token does not, a pattern crypto investors know well and one worth recalling before treating any of these tokens as a proxy bet on the sector’s growth.

What resolution looks like, and what it means for the machines

Three broad outcomes are on the table, and each points the agents in a different direction. If the Supreme Court sides with Kalshi and federal preemption wins, a single CFTC license becomes a national passport. Agents keep trading a unified, deep, 50-state market, institutions pile in, and the perps-and-Pro build-out accelerates. If the Court sides with the states, the market fragments along state lines, geoblocking gets serious, and the cross-jurisdictional convenience that makes agent deployment so cheap turns into a liability instead. And if the CFTC’s Rule 40.11 rewrite survives its 44-state gauntlet while the classification fight grinds on, the likeliest near-term reality is a patchwork: sports contracts contested state by state while other event contracts trade freely, with agents concentrating wherever the legal ground is firmest.

For anyone building or deploying these agents, the practical takeaways are unglamorous. Treat legal jurisdiction as a live input to the strategy, not a footnote. Assume the operator, not the software, wears the liability, and structure accordingly. Do not confuse a framework’s traction with its token’s price. And watch two dockets and one rulemaking more closely than any order book, because in 2026 the biggest variable in a prediction-market agent’s expected value is not its model; it is which court gets the last word. The machines learned to trade faster than almost anyone expected. Now they are waiting, like everyone else, to find out whether they are allowed to.

Frequently Asked Questions

Are prediction markets legal in the United States in 2026?

It depends on the contract and, increasingly, on the state. Event contracts on a CFTC-registered exchange such as Kalshi operate under federal law, but an August 2026 Ninth Circuit ruling confirmed that states can still treat sports contracts as gambling, and several states have sued. Non-sports event contracts are generally less contested than sports ones, and the Supreme Court is now likely to settle the core jurisdiction question.

Who regulates prediction markets, the SEC or the CFTC?

The CFTC. Prediction-market contracts are event contracts under the Commodity Exchange Act, so the Commodity Futures Trading Commission is the primary federal regulator, and it is currently rewriting Rule 40.11. The SEC’s role is limited to the crypto tokens attached to the ecosystem, such as UMA or Olas, not the wagers themselves. State gaming regulators are the third force fighting for authority.

How much of prediction-market trading is done by AI agents?

By the middle of 2026, more than 30% of active wallets on Polymarket were autonomous AI agents, according to analytics cited by CoinDesk. Agents also outperform people: roughly 37% post positive returns versus 7% to 13% of humans, and 14 of the 20 most profitable wallets on the platform belong to bots rather than human traders.

What is the CFTC Rule 40.11 rewrite about?

It replaces a blanket ban on event contracts tied to certain enumerated activities (terrorism, assassination, war, gaming, and unlawful conduct) with a three-step, case-by-case test of whether a contract involves such activity and is contrary to the public interest. Proposed in June 2026, it drew a joint objection from 44 state attorneys general who argue the agency overstepped its statutory authority.

Can an AI trading agent be held liable if it breaks the law?

No, not the agent itself. Software has no legal personhood, so it cannot hold a license or be fined. Liability instead falls on a human or company, the deployer, operator, or developer, depending on who controlled the conduct. That accountability gap is one of the hardest open questions as autonomous agents come to dominate these markets.

Marcus Okafor is a senior markets writer at HOGE Wire covering the intersection of crypto, AI, and financial regulation.