The Glass Market: Prediction-Market Insider Trading in 2026

In February 2026, Polymarket listed one of the odder contracts in its history: a market on which crypto company the on-chain investigator ZachXBT would next expose for insider trading. Within days it had traded around $40 million, and then the obvious happened. Before ZachXBT published a word, the odds lurched toward one name, Axiom. A wallet called “predictorxyz” turned an average entry near $0.14 on 477,415 shares into about $411,000, roughly a sevenfold return, while at least a dozen wallets split more than $1 million between them, according to CoinDesk. The market built to catch insider trading had itself been insider-traded.

That episode is a near-perfect miniature of what prediction markets became in 2026. They are the most transparent trading venues in finance: on Polymarket, every position, every wallet and every fill sits on a public blockchain for anyone to read in real time. That radical visibility is exactly why autonomous AI agents came to dominate them, and it is why they turned into the most heavily watched markets in crypto. It is also why, paradoxically, they may be the easiest venue in modern finance for someone holding real-world information to cash in. Transparency, it turns out, detects insider trading beautifully. It does almost nothing to prevent it.

By September 2026 that paradox had stopped being a debating-society topic. The CFTC’s enforcement chief called the notion that insider rules do not apply “a myth”; the Justice Department and the CFTC brought their first prediction-market insider cases; WIRED pried loose records of three quiet federal probes; and Polymarket referred around 100 wallets to law enforcement. This is the story of the glass market: who is watching, who is trading on what they should not, and why the machines on both sides make it so hard to tell the two apart.

The market that watches itself

Start with the plumbing, because the plumbing is the whole argument. Polymarket settles on Polygon, an Ethereum scaling network, which means its order book and its resolved positions are public data. Anyone with a block explorer, an Arkham dashboard or a Dune query can watch a wallet build a position in real time and read its full history afterward. Kalshi is different in form, a CFTC-registered designated contract market with know-your-customer checks and a centralized matching engine, but its large positions and flows are tracked almost as closely by a cottage industry of analysts. The public price on these venues has become a machine-readable feed that other software reads and reacts to, the same way an automated market maker’s curve becomes a de facto oracle.

The consequence is a property no traditional exchange shares. On Nasdaq you see an anonymized, aggregated tape; you cannot point at the specific account that just bought. On Polymarket you can point at the exact wallet. The insider, if there is one, leaves fingerprints on a public ledger. The problem is that the wallet is pseudonymous and, on the offshore main platform, there is no identity check at the door. Maximum visibility, minimum gatekeeping: that is the glass market in a single line, and everything that follows, the surveillance economy, the enforcement cases and the definitional fights, flows from it.

How the machines took the order book

Agents did not sneak into prediction markets; they took over the busiest one. CoinDesk reported in March that more than 30% of wallets on Polymarket were running AI agents, and that the agents were winning: around 37% of the agents behind Valory’s Polystrat product showed positive profit and loss, against only 7% to 13% of human traders. Polystrat alone executed more than 4,200 trades in roughly a month, with a single-trade return that touched 376%.

David Minarsch, chief executive and co-founder of Valory AG, is blunt about what that means. “In a nutshell, Polystrat is an autonomous AI agent that trades on Polymarket 24/7 on behalf of its human user,” he told CoinDesk, adding that “humans are already in a battle with machines” and that “agents tend to do better than humans.” He is equally blunt about the failure mode: “Simply prompting off-the-shelf models with markets usually results in outcomes no better than a coin-flip.” The edge, in other words, is not the model; it is the workflow wrapped around it, and increasingly the data pipeline feeding it.

That pipeline is the public chain, which is where transparency turns treacherous. A study of Polymarket’s order-book data by researcher Philipp Dubach found that inferring the direction of a trade from the public feed agrees with the on-chain truth only about 59% of the time, well below the roughly 80% that standard methods achieve on Nasdaq, because the visible book is thin, near-uniform in depth and delivered with a sub-50-millisecond median latency that hides a multi-second tail. A naive bot reading the glass can get order flow backwards. The same agents now also trade the perpetual futures both venues have added, a market whose mechanics HOGE Wire covered in its guide to on-chain perpetuals.

When the insider-detection market got insider-traded

Return to the ZachXBT market, because it is the cleanest illustration of the paradox. The contract asked which company the investigator would name in an upcoming report on insider trading. For most of the week the favorite was a rival name, Meteora, trading above 50%, with Axiom trailing. Then, late on the Wednesday before publication, the odds swung toward Axiom and peaked around 46%. When ZachXBT posted his findings on Thursday morning, Axiom was indeed the answer, and the wallets that had front-run the swing were paid.

The numbers, via CoinDesk, are almost comic. Roughly $40 million had changed hands since the start of the week. The wallet “predictorxyz” turned an average entry near $0.14 into about $411,000. A separate cluster of five wallets staked around $50,000 and collected some $266,000. ZachXBT, to his credit, had contacted Axiom for comment before publishing, which made a leak, as he put it, probably inevitable. That is the point: the information existed off-chain, in a handful of inboxes, before it existed on-chain, and anyone in that loop could bet on a venue that asks no questions. A market designed to expose insiders became a case study in how they operate.

“Insider trading just accelerates the truth”

If radical transparency makes insider activity visible, someone was always going to build a business on reading it. That someone is Polysights, an AI-powered analytics service founded by Tre Upshaw, a 29-year-old former memecoin trader from Canada. Polysights scores trades across roughly eight signals, including how large the wager is, how new the wallet is, how long the odds were when the trader entered, and how concentrated the trader’s activity is in one or two markets, then flags the patterns that look like informed money. It has amassed around 24,000 users and raised about $1.5 million, backed in part by Polymarket itself, per CNBC.

Upshaw’s own framing is the philosophical heart of the debate. “Insider trading just accelerates the truth faster at the end of the day,” he told Bloomberg, which used his data for an investigation that reviewed some 34,000 flagged transactions between August 2025 and June 2026, much of it clustered in geopolitical markets tied to Iran and Venezuela. Bloomberg estimated that roughly $200 million in first-half 2026 trades carried the fingerprints of possible informed activity.

Transparency does not stop there. Because the suspected-insider wallets are public, other traders simply copy them. Apps that let ordinary users mirror the positions of Polymarket’s sharpest, and most suspicious, wallets went viral in early 2026 under a blunt slogan: if you cannot beat the insiders, join them. The obvious caveat, which honest analysts stress, is that a flag is a suspicion, not a verdict; a wallet that always seems to know may be a genuinely sharp forecaster, and Polymarket’s own analysts have pushed back on specific Bloomberg flags as ordinary bettors caught by a noisy filter.

The myth the CFTC killed

For years a folk belief circulated among prediction-market traders that insider rules simply did not reach event contracts, that betting on information was categorically different from trading a stock on it. In 2026 the CFTC set out to kill that belief by name. Speaking at New York University’s law school on March 31, David Miller, the agency’s director of enforcement, warned that “a myth has spread that insider trading is permissible, even encouraged, in prediction markets,” and named insider trading in prediction markets a top priority for his division, according to a summary by Morrison Foerster.

The legal machinery he pointed to is not new. The CFTC treats event contracts as swaps under the Commodity Exchange Act, which brings them under Section 6(c)(1) and Rule 180.1, anti-fraud provisions added by the Dodd-Frank reforms and modeled directly on Section 10(b) and Rule 10b-5 of the securities laws. Those provisions import a misappropriation theory of insider trading into the swap markets. The agency spelled out the reach in a February enforcement advisory, citing examples such as a candidate trading on his own campaign and a video editor trading channel contracts on nonpublic information, and framing liability around the “misappropriation of confidential information in breach of a pre-existing duty of trust and confidence to the source,” per the CFTC.

One point of US framing is worth stating plainly, because it is routinely muddled: this is the CFTC’s arena, not the Securities and Exchange Commission’s. Event contracts are derivatives under the Commodity Exchange Act, so venues like Kalshi and Polymarket’s US entity answer to the CFTC. The SEC enters only around the edges, through the crypto tokens that sit alongside these markets, such as UMA or OLAS, which is a separate securities question with its own answers.

The first cases

The theory stopped being abstract in May, when the Justice Department and the CFTC charged a Google software engineer who traded under the handle “AlphaRaccoon.” According to law firm Davis Wright Tremaine, he used internal Google tools to obtain confidential data about the company’s 2025 “Year in Search” rankings, then traded more than 20 event contracts tied to those rankings before their public release, hitting near-perfect accuracy and pocketing about $1.2 million between October and December 2025. The civil charges cite Section 6(c)(1) and Rule 180.1; the criminal charges add commodities fraud, wire fraud and money laundering.

A second case had already landed. Earlier in 2026, prosecutors in the Southern District of New York charged a US Army soldier with insider-trading-style offenses for placing prediction-market bets based on classified information about a planned military operation in Venezuela. The through-line matters: both cases rest on misappropriation, on information stolen or misused in breach of a duty, not on the mere fact that the trader knew something. That distinction is the fault line the rest of the fight runs along.

Three quiet probes

Enforcement is running well ahead of the headlines. In September, WIRED reported, through a Freedom of Information Act request, that the CFTC had authorized at least three previously undisclosed investigations into potential insider trading on Polymarket, approved between May and July 2026 and tied to markets on Joe Biden’s pardons, on Iran-related events, and on Google’s “Year in Search” rankings, per a writeup at Yahoo Finance. Chair Michael Selig personally approved the first, on the pardon markets, in early May, after one trader reportedly made more than $300,000 correctly betting on who would receive preemptive pardons. An internal CFTC email said investigators were examining additional individuals, and the Southern District of New York was running a parallel probe.

The political layer is loud, too. Lawmakers have pressed the agency to investigate suspicious Polymarket trades that preceded a US-brokered Iran ceasefire announcement, arguing the pattern looked like someone trading ahead of the news. And the venues are feeding the pipeline themselves: Polymarket’s chief legal officer, Neal Kumar, said the company had referred nearly 100 wallets to law enforcement after its surveillance flagged them, per crypto.news, while CNN reported in August that it had sent dozens of possible military-related cases to the Justice Department. Transparency, after the fact, is a prosecutor’s dream.

Is it even insider trading?

Here is where the glass market gets genuinely hard, and where reasonable lawyers disagree. Securities insider-trading law is built on a relationship: a corporate insider, or someone who misappropriates an issuer’s material nonpublic information in breach of a duty, trades that company’s securities. Prediction markets frequently have none of those parts. There is no issuer, no share, often no fiduciary. What is the “inside” of a contract on whether it will rain in Miami, or on who wins an election?

The CFTC’s answer, the misappropriation theory, works cleanly when the information was stolen: the Google engineer breached a duty to his employer, the soldier misused classified intelligence. It gets murky fast for the trader who simply knows something legitimately, the reporter about to publish, the staffer who lawfully read a memo, the local who can see the storm coming. Prediction-market theory actively wants those people trading, because the entire design goal is to pull dispersed private information into an accurate public price. Upshaw’s line, that insider trading accelerates the truth, is not a confession so much as a restatement of the academic case for these markets. The same behavior is a feature to the market designer and a felony to the prosecutor, and which one it is turns entirely on where the information came from.

That is a shaky foundation for a market this large, and watchdogs have noticed. The Project On Government Oversight has argued that existing law is a patchwork and that Congress should write a clear insider-trading standard for prediction markets rather than leaning on swap-market case law built for a different instrument, per its fact sheet. The CFTC’s separate rewrite of its event-contract rule governs which contracts are allowed to list, not who may trade them, so for now the conduct question lives with Section 6(c)(1) and Rule 180.1, one case at a time.

Detection is not prevention

If the chain sees everything, why is any of this still happening? Because seeing is not stopping. The offshore Polymarket platform performs no identity check at signup, wallets are pseudonymous, and the information asymmetry that drives the trade lives in the real world, not on the ledger. The chain records the bet; it cannot see the leaked inbox that motivated it. By the time an analytics engine flags a pattern, the position is filled, the market has resolved and the winnings can be bridged elsewhere.

So it becomes cat and mouse. Polymarket rewrote its market rules in early 2026, tightened surveillance and delisted some of the riskiest contracts. Informed traders responded the way informed traders always do: fresh wallets for each play, orders split to look ordinary, entries timed at plausible odds to slip under an eight-metric score. Attribution, the act of tying a pseudonymous wallet to a human being, is the real bottleneck, and it usually comes down to a subpoena to a centralized exchange or an operational-security slip, the kind of leak HOGE Wire has examined in the context of mobile wallet risk. The frictionless onboarding that helps ordinary users, a topic covered in our look at wallet UX, also lowers the bar for the people the rules are meant to stop.

Machines watching machines

Layer the agents back on and the picture turns reflexive. The trading side is automated: bots that read the public feed in milliseconds, split and route orders, and chase the long tail of thin markets that Minarsch calls the most interesting territory for AI. The watching side is automated too: Polysights is, functionally, a surveillance agent that scores every wallet on the same public data the traders are mining. One set of machines tries to profit from information; another set tries to flag it. They read the identical ledger.

The loop feeds on itself. When a surveillance tool or a copy-trading app highlights a suspicious wallet, other bots pile into the same position within blocks, which crowds the trade, moves the price and blurs the very signal that raised the alarm. And because so much of the flow now runs on a handful of frontier models, the agents increasingly cluster on the same reads, which thins everyone’s edge. A controlled study that ran six frontier models on real prediction markets found they mostly lost money on Kalshi and roughly broke even on Polymarket, a reminder that being informed and being profitable are not the same thing, and that a market crowded with correlated machines can be efficient and treacherous at once, as one benchmark laid out. The wisdom of crowds, on the glass market, is increasingly the wisdom of software watching other software.

The resolution blind spot

One more surface deserves a mention, because insider activity is not the only place transparency and manipulation meet. Someone still has to decide who won. Polymarket resolves through UMA’s optimistic oracle, where a proposer posts an outcome and, if it is disputed, the question escalates to a token-holder vote. In 2026 UMA tightened the proposer side with a managed-proposer whitelist, requiring at least five proposals over a rolling six months at 95% accuracy or better plus a $750 bond, and it reports a dispute rate near 1.3% against usage that has grown almost 500% in a year, per its own writeup. That reduces frivolous proposals, but it governs who may propose, not who votes on a contested outcome, and the voting layer carries its own conflict-of-interest questions. HOGE Wire dug into that machinery separately in Bots Bet, Machines Judge; the short version is that the settlement layer is one more place where a public, incentive-driven process can be gamed by whoever shows up with the most at stake.

What the numbers say

For all the turbulence, this is a large and institutionalizing market. Combined monthly volume at Kalshi and Polymarket reached $45.33 billion in August 2026 even as it fell 14.5%, the first month-over-month decline in a year, with Kalshi at $37.17 billion and Polymarket at $8.16 billion, per The Crypto Basic. The New York Stock Exchange’s parent, ICE, has committed on the order of $2 billion to Polymarket on a data-and-infrastructure thesis rather than a betting one, per CoinDesk, and Bernstein has projected the sector could reach $1 trillion by 2030, per CNBC.

The token layer tells a quieter story, and it is the part the SEC would care about. UMA, the asset that secures Polymarket’s oracle, traded around $0.38 for a market capitalization near $34 million, ranked outside the top 600, per CoinGecko. OLAS, the token tied to the Valory and Olas agent stack, sat near $0.028 for a capitalization around $8 million. Even the entire AI-agents token category, at roughly $3.45 billion, is a rounding error next to the volume these venues clear. The lesson repeats across crypto: the quality of the venue or the framework is not the price of its token.

The perimeter nobody owns

Step back and the glass market has exposed a governance gap that no single body fills. The venues can surveil but not prosecute. The CFTC can prosecute but reaches only misappropriation, and only in the United States. A growing list of states insists the whole category is unlicensed gambling and wants it under their own rules. And Europe has drawn a harder line: regulators led by ESMA have warned that neither Polymarket nor Kalshi holds authorization anywhere in the bloc, and several member states have moved to block them, a supervision-and-enforcement problem HOGE Wire traced in its coverage of MiCA in 2026 and, at the EU level, in ESMA’s standing guidance on binary-outcome contracts.

The AI layer sharpens every edge of that gap. When roughly a third of the flow is autonomous, and the watchers are autonomous too, market abuse becomes a machine-versus-machine problem that insider law written for human fiduciaries was never designed to handle. Transparency made these markets legible, tradable and surveillable; it did not make them clean. Detection scaled; deterrence lagged. Whether 2026’s cases actually change behavior will come down to two unglamorous things: attribution, the slow work of linking wallets to people, and a political decision about what “insider” even means when the market is public, the traders are pseudonymous, and half of them are software.

Frequently Asked Questions

Is insider trading illegal on prediction markets like Polymarket and Kalshi?

The CFTC says yes. It treats event contracts as swaps under the Commodity Exchange Act, so trading on misappropriated material nonpublic information can violate Section 6(c)(1) and Rule 180.1. Enforcement director David Miller called the idea that insider rules do not apply “a myth,” and in 2026 the Justice Department and CFTC brought their first cases. The catch is that liability generally requires the information to have been obtained in breach of a duty, so merely being well informed is not automatically illegal.

How is insider trading detected on Polymarket?

Because Polymarket settles on a public blockchain, every wallet and position is visible. Analytics services such as Polysights score trades across signals like wager size, account age, entry odds and market concentration, then flag patterns that look like informed money. Bloomberg reported roughly 34,000 flagged transactions and about $200 million in suspicious trades in the first half of 2026, though a flag is a suspicion, not proof.

Which US regulator oversees prediction markets, the CFTC or the SEC?

The CFTC. Event contracts are treated as swaps or derivatives under the Commodity Exchange Act, which places venues like Kalshi, a designated contract market, and Polymarket’s US entity under CFTC jurisdiction. The SEC only enters through the crypto tokens that sit alongside these markets, such as UMA or OLAS, which is a separate securities question.

Do AI agents really dominate prediction-market trading?

On Polymarket, largely yes. CoinDesk reported that more than 30% of wallets use AI agents, and that around 37% of the agents behind Valory’s Polystrat showed positive returns versus 7% to 13% of human traders. Valory chief executive David Minarsch describes Polystrat as an autonomous agent that trades 24/7, while warning that naive off-the-shelf models perform no better than a coin-flip.

Why does on-chain transparency not stop insider trading?

Transparency exposes trades after they happen, but it does not gate who can trade. The main Polymarket platform is offshore with no identity checks at signup, wallets are pseudonymous, and the underlying information leaks in the real world rather than on the chain. So watchers can flag suspicious wallets quickly, yet linking a wallet to a person, and proving the information was misused, stays slow and hard.

By Marcus Okafor, senior markets writer at HOGE Wire, covering the collision of AI agents, on-chain markets and financial regulation.