Prediction Markets Have an Insider Trading Problem That Market Theory Can't Fix

Two days ago, the Department of Justice unsealed an indictment¹ against Gannon Ken Van Dyke, a U.S. Army Special Forces master sergeant who allegedly used classified information about Operation Absolute Resolve — the military operation that captured Venezuelan leader Nicolás Maduro — to place 13 bets on Polymarket and pocket over $400,000. He faces up to 20 years in prison. The CFTC filed parallel civil charges² the same day, marking the agency's first-ever insider trading enforcement action against a prediction market participant.

I want to take this case seriously. Not as an isolated incident of a rogue soldier, but as the first confirmed proof-of-concept for something that critics of prediction markets have warned about for years: that these platforms, designed to aggregate dispersed knowledge into probability estimates, have become a mechanism for monetizing privileged access to classified information. And I think the evidence now available shows that this problem is structural, not incidental — meaning no amount of enforcement against individual bad actors will fix it without deeper changes to how these markets operate.

Here's why the Van Dyke case matters more than it appears. The indictment tells us Van Dyke created his Polymarket account on December 26, 2025, and placed all 13 bets between December 27 and January 2. Every single bet took the "YES" position on U.S. forces entering Venezuela or Maduro being removed. He was, according to ABC News³, "believed to be the first instance of the Department of Justice prosecuting a case of insider trading on a prediction market." But he is very clearly not the first person to have done it. He's just the first one dumb enough to get caught in an obvious way — he tried to delete his account and move his profits to a brokerage in his own name after seeing media reports about suspicious trading.

The pattern is now well established. Before the February 28 U.S.-Israeli strikes on Iran, six newly created Polymarket accounts purchased "Yes" shares at roughly ten cents apiece⁴ hours before the first explosions hit Tehran, collectively earning about $1.2 million. A trader called "Magamyman" made $553,000⁵ betting on Khamenei's death seventy-one minutes before the news broke. In April, at least 50 brand-new accounts⁶ placed bets on a U.S.-Iran ceasefire in the hours and minutes before Trump announced it on social media. Israeli authorities have indicted two individuals⁷, including a military reservist, for using classified information to bet on Polymarket during Israel's war with Iran. And a CNN investigation⁷ found one trader who won 93% of their five-figure Iran-related wagers, earning nearly $1 million since 2024.

These aren't anecdotes. Researchers at Columbia Law School and the University of Haifa recently published the first systematic analysis⁸ of potentially informed trading on Polymarket, screening over 93,000 distinct markets and 50,000 wallet addresses between 2024 and 2026. They found 210,718 suspicious wallet-market pairs with a 69.9% win rate — "a result that exceeds the null distribution of random chance by more than 60 standard deviations." Their estimate of aggregate anomalous profit: approximately $143 million. Co-author Joshua Mitts of Columbia Law told NPR⁹ that some individual trading patterns had odds of occurring by chance that were "virtually zero."

Now, prediction market advocates have a strong counter-argument, and I want to engage it honestly. The theoretical case for these markets — first formalized by Hayek and operationalized in Robin Hanson's work — is that paying people to be right about the future elicits better information than asking them politely. Philip Tetlock's Good Judgment Project, part of IARPA's Aggregative Contingent Estimation program, showed that crowd-aggregated forecasts outperformed intelligence analysts with classified access by about 30% on calibration metrics. When Polymarket called the 2024 election more accurately than 538 and most polling aggregators, that was a genuine vindication of the mechanism. And an academic paper analyzing Polymarket's 2024 election data¹⁰ found that as trading volume grew, Kyle's lambda (a measure of manipulation vulnerability) declined by more than an order of magnitude, suggesting the market became harder to manipulate as it matured.

These are real findings. I don't dismiss them. But I think they answer the wrong question. The issue is not whether prediction markets can produce accurate probability estimates under favorable conditions. They can. The issue is what happens when the mechanism is deployed at financial scale in domains where outcomes are determined by classified government decisions, and where the people with advance knowledge have a direct financial incentive to trade.

The gap between the IARPA experiment and today's prediction market reality is enormous. The ACE program measured disinterested forecasters using only public information, with small financial stakes. Polymarket in 2026 is a $15 billion-valued platform where, according to Bernstein¹¹, total market volumes are projected to reach $240 billion this year alone — a 370% increase over 2025 — with $1 trillion projected by 2030. Combined year-to-date volume on Kalshi and Polymarket has already exceeded $60 billion¹², surpassing all of 2025. These numbers mean the incentive to exploit insider information scales proportionally. A soldier with operational knowledge of a planned military strike can now convert that information into six-figure profits with a few clicks on a phone.

The regulatory response, while accelerating, is fundamentally mismatched to the speed of market growth. The CFTC issued its first prediction markets advisory¹³ in February 2026. Director of Enforcement David Miller declared at NYU¹⁴ on March 31 that insider trading on prediction markets is a "top priority" and called the idea that insider trading law doesn't apply to these markets "a myth." The CFTC published an advance notice of proposed rulemaking¹⁵ on March 12 seeking public comment on event contract regulation. Congress has at least four bills pending¹⁶ addressing prediction market insider trading. All of these are necessary steps. But notice the timeline: the CFTC's first enforcement action arrived in late April 2026, more than three months after the Iran strikes generated $529 million in trading volume and multiple documented episodes of suspicious activity. The agency is building the fire truck while the building is already on fire.

There's a deeper structural problem that regulation alone may not solve. Traditional insider trading in equities is hard enough to prosecute, and there the SEC has had decades to develop surveillance tools calibrated to markets with continuous price anchors like earnings reports and economic data. Political and military event contracts have no such anchor until resolution. As the Congressional Research Service noted¹⁶, the legal framework for prosecuting prediction market insider trading is built on the misappropriation theory — you can only violate the law if you breach a duty of trust and confidence to the source of the information. A soldier trading on classified ops clearly breaches that duty. But what about a White House staffer's cousin who overhears something at a dinner? What about a foreign intelligence officer who gleans information about U.S. military planning and trades on an offshore platform that doesn't collect identity information? Polymarket's international exchange, which handles the vast majority of volume, doesn't conduct identity checks.

And then there is the conflict-of-interest elephant in the room. Donald Trump Jr. serves as an adviser to both Polymarket and Kalshi¹⁸. The CFTC's chair, Brian Quintenz, was nominated by Trump and previously served on Kalshi's board¹⁹. The Trump administration's policies — unpredictable by design, announced via social media — are the single largest driver of prediction market volume. As Virginia Tech economist Kwok Ping Tsang put it to Fortune¹⁸: "Trump is the guy. He makes the market possible." When the president's family has financial interests in the platforms that profit from his unpredictability, and 50 anonymous accounts bet on a ceasefire minutes before he announces it, the epistemic integrity of the entire mechanism comes into question.

I am not arguing that prediction markets should be abolished. Their forecasting accuracy under clean conditions is genuinely impressive, and the transparency of blockchain records makes abuse at least detectable in ways that corruption in traditional intelligence or policymaking channels is not. But I think the current trajectory — explosive volume growth outpacing regulatory capacity, geopolitical contracts traded on platforms with no identity verification, and documented insider exploitation at multiple points in the system — means these markets are functioning less as truth aggregators and more as intelligence bazaars.

The indicator to watch is straightforward: does the CFTC's enforcement pipeline scale with market growth, or does it remain a handful of prosecutions per year against the most obvious offenders while the $143 million in anomalous profits continues to accumulate? If, twelve months from now, the Van Dyke case is still the only criminal prosecution while prediction market volume has doubled again, we'll have our answer about whether the regulatory apparatus can keep pace. My prediction: it can't, and the industry's enthusiasts will continue to treat every detected abuse as proof of the system's resilience rather than evidence of its vulnerability.