AI agents are now placing trades on prediction markets through the same APIs that human developers use, and regulators are scrambling to keep pace. Platforms like Kalshi and Polymarket have built developer tools that let automated systems authenticate, read order books, and execute bets without any manual input. As the CFTC’s Enforcement Division issues new warnings about misconduct in event contracts, the collision between machine-speed trading and a still-unsettled legal framework is raising hard questions about who bears responsibility when software, not a person, pulls the trigger on a wager.
How Platforms Built the On-Ramp for Bots
Both Kalshi and Polymarket have invested heavily in making their markets accessible to automated systems. Kalshi’s developer portal provides REST and WebSocket interfaces, SDKs, an OpenAPI specification, and a demo environment, all designed so that AI agents can stream orders and access real-time market data without touching a browser. The platform exposes orders, trades, portfolio history, and public order books through these endpoints, making it straightforward to plug in an automated strategy.
Polymarket takes a similar approach. Its documentation describes a Gamma API for market metadata and a CLOB API for orderbook and pricing data, enabling agent systems to fetch market lists, compute probabilities, and route orders programmatically. The platform also maintains an official Python client, py-clob-client, hosted on GitHub, which walks through how automated traders authenticate with a private key, fetch order books, and place trades on the central limit order book. Polymarket’s ecosystem programs for market makers and builders further signal that the platform actively courts algorithmic participants who can add depth and liquidity.
This infrastructure matters because it lowers the barrier to entry for anyone building an AI trading agent. A developer with basic Python skills and an API key can deploy a bot that monitors dozens of event contracts simultaneously, identifies price discrepancies, and executes trades in milliseconds. The result is a market where the fastest participants are not humans scanning headlines but software scanning endpoints.
As large language models gain tools for calling external APIs, the line between “bot” and “assistant” blurs. A conversational agent that can read market data and submit orders can be embedded inside messaging apps, trading dashboards, or even news sites. Once connected, these agents can react to new polls, court decisions, or economic data far faster than a human, updating positions in a continuous feedback loop that few retail traders can match.
The CFTC Draws a Line on Misconduct
Speed and automation do not exist in a regulatory vacuum. The CFTC’s Enforcement Division issued a prediction markets advisory that lays out how the agency views misconduct risks in event contracts and applies specific enforcement theories to this asset class. The advisory summarizes two enforcement matters tied to trading on KalshiEX, including one involving a person who had influence over the outcome of the event being traded. That case is a direct warning: if someone with the power to affect a result trades on that result, the CFTC considers it a form of market abuse analogous to insider trading in traditional commodities.
The agency has already brought high-profile actions in this space. It ordered Blockratize, Inc., doing business as Polymarket, to pay a civil penalty after finding that the platform offered event markets beginning around mid-2020 without proper registration. As part of that settlement, Polymarket was required to wind down non-compliant markets and bring its operations into line with federal derivatives rules.
What most coverage misses is how automation compounds these risks. When a single human trader with inside knowledge places a bet, the misconduct is contained. When that same knowledge is fed into an AI agent that can split orders across multiple accounts, time entries to avoid detection, and optimize position sizing, the scale of potential abuse grows exponentially. An agent can also be configured to trade across multiple platforms at once, making it harder for any single venue to detect unusual patterns.
The CFTC’s advisory does not explicitly address AI agents, but its enforcement theories around outcome-influencing traders apply regardless of whether the order originates from a keyboard or an algorithm. In practice, that means responsibility will likely fall on the person or entity that deploys the agent, even if the specific trading decisions are made autonomously. For platforms, the presence of bots raises the stakes for surveillance: monitoring for collusion, wash trading, and manipulation now requires tools that can recognize algorithmic fingerprints as well as human behavior.
Election Contracts and the Legal Tug of War
The regulatory picture is further complicated by an unresolved federal dispute over election event contracts. The CFTC attempted to block Kalshi’s congressional election markets, resulting in the case known as Kalshi v. CFTC. An appeals court later lifted an injunction that had stopped trading, allowing contracts on U.S. congressional elections to resume temporarily. That decision did not settle the underlying question of whether such markets are permissible under federal law; it simply removed a preliminary barrier.
The core legal fight centers on whether election contracts are off-exchange event-based derivatives that run afoul of bans on gaming and unlawful options, or whether they are bona fide hedging tools that should be allowed under carefully crafted rules. During the appeals process, judges have already heard arguments about the scope of the CFTC’s authority and the potential impact of allowing or prohibiting political prediction markets. While the court deliberates, Kalshi has experienced interim trading windows in which election contracts are live, then paused, then potentially revived again.
This uncertainty creates a peculiar environment for automated trading. AI agents do not pause for legal ambiguity. If an API endpoint is live and contracts are listed, a bot will treat them as tradeable. The interim windows that courts create while deliberating become opportunities for high-frequency strategies that can enter and exit positions before a ruling lands. Human traders, who need time to read filings and assess legal risk, operate at a structural disadvantage during these periods.
The legal limbo also complicates risk management. An agent programmed to maximize expected value might ignore the possibility that a sudden court decision could cancel markets or restrict withdrawals. Developers who fail to encode these legal contingencies into their bots could find their systems holding large positions in contracts that disappear overnight, or trapped on platforms that must rapidly reconfigure their offerings to stay compliant.
Federal Support, State Bans, and a Jurisdictional Maze
Adding another layer of tension, the Trump administration has signaled sympathy for platforms like Kalshi and Polymarket even as some state authorities move to restrict or ban prediction markets. Sports-related volume also plays a role in the overall composition of these platforms, blurring the line between event betting and traditional gambling in ways that complicate state-level responses. To some regulators, election and policy markets look like derivatives; to others, they resemble online wagering that should be governed under gambling statutes.
The resulting federal-state conflict means that a bot operating legally under federal rules could simultaneously violate state law, a jurisdictional mess that no automated system is equipped to sort out on its own. An AI agent that trades from servers in one state on a platform registered elsewhere, on events tied to yet another jurisdiction, raises difficult questions about which laws apply. For now, most platforms push that responsibility onto users through terms of service, but automation makes it easier for traders to ignore or misunderstand those boundaries.
Why Arbitrage Agents Could Flatten Global Odds
One consequence of widespread API access that deserves more scrutiny is the effect on price discovery across fragmented markets. When AI agents can simultaneously monitor Kalshi’s REST endpoints, Polymarket’s CLOB, and potentially offshore venues, they can identify and exploit price differences in near-real time. A contract trading at 62 cents on one platform and 58 cents on another represents a profit opportunity for a bot fast enough to buy low and sell high before prices converge.
At scale, this arbitrage activity tends to homogenize probabilities across platforms. If enough agents are scanning for gaps, a mispriced contract will be quickly corrected as bots pile in on the cheap side and sell on the expensive side. That can improve informational efficiency: no single market can drift too far from the consensus without being disciplined by cross-venue traders. For users, it means that odds on major political or economic events may look nearly identical whether they are viewing a regulated U.S. exchange or a crypto-native prediction site.
But there are trade-offs. When arbitrage agents dominate, local idiosyncrasies in pricing (caused by different user bases, information flows, or regulatory constraints) get ironed out. Smaller platforms may struggle to attract liquidity if their prices are instantly mirrored elsewhere. And in moments of stress, the same algorithms that keep prices aligned can transmit shocks globally, as bots rush to unwind positions across all linked venues at once.
For regulators, this interconnectedness raises systemic questions. An enforcement action or sudden rule change on one platform could ripple through others if arbitrage agents are forced to close positions or shift volume. Conversely, a manipulation scheme executed by an AI agent on a lightly regulated offshore market might subtly influence odds on a registered U.S. exchange via cross-market trading, even if that exchange has robust internal controls.
Prediction markets were once niche websites where hobbyists bet small sums on elections and sports. With industrial-grade APIs and AI agents now in the mix, they are starting to resemble a global, always-on information layer about future events, one that moves at machine speed, but is still governed by laws written for a slower era. How regulators, platforms, and developers respond to that mismatch will determine whether automated prediction trading becomes a durable part of the financial landscape or a brief experiment cut short by legal and ethical backlash.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
