When Prediction Markets Become Weapons
How Institutional Prediction Markets Create New Vectors for Misinformation

Prediction markets have exploded in popularity in recent years and, like in the case of many emerging technologies, have rapidly encountered external pushback. Major industry players such as Polymarket and Kalshi have faced regulatory challenges from states over their sports betting offerings and, in Polymarket’s case, have received flak over insider and wash trading. To match this external pressure, the vast majority of the industry’s attention has been concentrated on these issues. Kalshi has dedicated 10-15 percent of its staff to market integrity, and Polymarket has partnered with Palantir to better track wallets.
But I think a more pressing issue has been neglected. In tandem with their commercial growth, prediction markets are increasingly being adopted by financial infrastructure firms and major news outlets as credible signals of public sentiment. This makes them a really attractive tool for misinformation campaigns and troll farms, which have historically struggled to get their narratives picked up by mainstream media. In this article, I’ll outline the components of modern prediction market structure that make them inherently vulnerable to targeted manipulation and ultimately demonstrate how such an attack might unfold.
I) The Theory
Before diving into potential vulnerabilities, I think it’d be beneficial to first establish what modern prediction markets aim to achieve and how a typical market works. Major players like Polymarket and Kalshi have long marketed themselves as platforms that aggregate "collective intelligence” and “harness the wisdom of crowds.” But what is this concept and where does it come from?
To understand this, let's look back at a 1904 Nature article, in which Francis Galton, a British statistician, observed a crowd of around 800 at a fair wager on the weight of an ox. His observation was that, while there were presumably many butchers, farmers, or experts in the crowd, there were also many bettors who had little to no “insider” knowledge on how to size up cattle. After everyone had submitted their guesses, Galton added all the contestants' estimates and calculated the mean of the group's guesses. That number represented the collective wisdom of the crowd. If the crowd were a single person, that was how much it would have guessed the ox weighed. The crowd guessed 1197, the ox weighed 1198 pounds[^1]. He deduced that when the guesses of a large, diverse crowd are aggregated, the errors cancel out and the “truth” emerges.
Later, in 1945, Friedrich Hayek, an Austrian economist, was frantically refuting prominent socialist economists such as Oskar Lange, who, after observing how Western governments successfully directed economies in World War II, advocated heavily for their centralization. Hayek instead believed that a centralized authority, regardless of how just and moral it were, could never fully collect or process the vast amount of highly localized information carried by millions of individuals regarding their specific industries and unique circumstances. In his paper The Use of Knowledge in Society, he argued that the world’s existing mechanism for setting prices, one that we stumbled upon naturally and accidentally, acted as a decentralized communications system. Prices automatically do the math that Galton did by hand, compressing scattered, human insights into a single metric.
Robin Hanson, the “father of prediction markets,” extended Hayek's logic from current economic conditions to future outcomes. Hanson reasoned that if market prices could efficiently aggregate dispersed information about the present supply and demand of commodities, they could do the same for future events. If you could create liquid markets on political elections or economic policy, the resulting contract prices would aggregate dispersed beliefs about those outcomes.
Now, when looking at how a binary market works, we can see a clear reflection of this theory in action:
Say you have a conjecture regarding the probability of X event happening. A contract trading at a $0.30 YES and a $0.70 NO means the market collectively believes there's a 30% chance of that outcome occurring. YES and NO contracts sum to $1, and your contract is worth $1 if X event happens, $0 if it doesn't. If you believe the true probability is higher, you buy, and if you're right, you profit $0.70 per share. The more confident you are, the more you buy and the more you shift the market in that direction as a result. In aggregate, those who believe an event has a higher chance of happening will buy, and those who believe otherwise will sell. This collectively nudges the price around; all valid information points the price in one direction, while erroneous guesses point in different directions and cancel each other out. This continues until the price converges and represents the collective opinion of the market.
II) Reality departs from theory.
The first problem in this application of the “wisdom of crowds” idea is that it assumes that all actors contribute meaningfully to price discovery. In practice, prediction market accuracy reflects more the wisdom of an informed minority rather than that of a crowd. The following paper by Gomez Cram et al. illustrates this. The researchers analyzed over 200 thousand markets on Polymarket and attempted to identify traders who were systematically profitable and thus drove prices closer to resolution value. They did this by repeatedly taking a given trader’s wallet, looking at the history of all their trades, and then randomizing whether that trader bought or sold a contract on that trade. If the trader’s actual profit and loss (PnL) was greater than this repeated, randomized PnL, that meant that the trader’s PnL couldn’t be explained by random chance and thus was systematically profitable.
Ultimately, these “skilled” winners made up just 3.14% of accounts. Lucky winners (positive returns indistinguishable from random chance) made up 29%; unlucky losers (losses indistinguishable from random chance) made up 61.4%; and unskilled losers (consistent loss in a statistically meaningful way) made up 6.4%. The remaining 0.1% were market makers.
An important distinction to make is whether or not these skilled traders included those who traded on inside information. The paper found that documented cases of insider trading typically fell within the lucky winners category, with only 1.1% of flagged insiders being classified as skilled. This makes sense as insider trading events tend to be large, one-off bets rather than consistent price discovery.
Furthermore, only the skilled group consistently predicted both the change in a contract's price in the following time period and the market's final resolution price. Around news releases like corporate earnings, they were the fastest at trading on public information the moment it was released. This means that the skilled group is the only group driving meaningful price discovery, and does so by accessing publicly available information faster than any other group.
The second problem is that information is often interdependent, meaning that market participants, rather than trading on their unique, independent market knowledge, often follow others' trades. Oftentimes, this takes the form of wallet tracking whales(traders who make up the top 1% of volume or are deemed to be consistently profitable, for instance). This 2025 Thesis paper analyzed US, Croatian, and German parliamentary markets and sorted Polymarket participants into tiers by trading behavior: whales, high-frequency traders, active traders, and retail traders, and then tracked how quickly each tier reacted after a whale changed position. In the 2024 US Presidential market, high-frequency traders moved within an average of 4.81 minutes of a whale trade, with active and retail traders trailing further behind. The author drew the following conclusion:
Rather than generating forecasts through the averaging of independent judgments, these markets operate through structured competition among specialized participants with different information access, analytical capabilities, and strategic approaches. Elite traders with superior information or analytical capabilities establish price trends that cascade through the market ecosystem, while other participants respond to these signals with varying levels of acceptance or resistance based on their own information sets and strategic positioning. This competitive dynamic can produce remarkably accurate forecasts despite—or perhaps because of—its hierarchical structure, as it enables sophisticated participants to exert disproportionate influence on price formation while creating financial incentives for information revelation through profit opportunities.
The unfortunate reality is that while prediction markets attempt to operate under a theoretical ideal, they succumb to normal market conditions. We see very clear parallels to the stock market. A select few drive price discovery, make most of the profits, and the rest either try to mimic the profitable or simply act as monetary cannon fodder.
The stock market functions fine under these circumstances, so why does the case differ for prediction markets? To understand this, we must look at depth and corrective trading. Market depth is how much money sits ready to absorb buy and sell orders before the price is shifted. A deeper market has more cash and thus requires more money to shift. Corrective trading describes what happens if a bad actor attempts to force a price in a direction away from its fundamental value. Rational actors are financially incentivized to trade against the deviation, and this trading activity is what pushes the price back to its correct value. While both prediction markets and stock markets observe these effects, they do so at vastly different scales. The stock market's impermeability to deliberate price tampering lies in the hundreds of billions of dollars in daily trading volume as well as the hundreds of institutional funds whose sole purpose is to profit off of minute inefficiencies. Prediction markets are devoid of both safety measures. They lack that level of depth and are comparatively bereft of large institutional presence, relying instead on individual traders to deploy high-frequency trading systems. This is the ultimate pitfall. There exists an opportunity for a well-resourced player to shift a market for a significant period of time and for an amount of capital that would never be achievable in a formal stock market.
While defenders of prediction market integrity point to cases like the 2024 election market, which had over $3.7 billion in trading volume, as evidence of how deep markets can get, the vast majority of markets never see this level of depth. The “Popular Vote Winner 2024," which ran alongside the main election market and asked which candidate would win the popular vote, ended with $628M in total lifetime volume. A trader with the alias “Théo” was able to, through many accounts, place multiple $500 thousand to $1 million bets on Trump winning and shift the contract's price from 26% to 39% in just a few hours. Théo was just an individual trader with high conviction on the outcome of this market, yet he still managed to move the market by 13 basis points. Important economic and political markets that touch on interest rates and political nominees range in depth anywhere from a couple hundred thousand to mid hundred-million-dollars, which, as we saw in the Theo case, are manipulable.
III) Problems become gateways for manipulation
This issue becomes more pressing because prediction markets still function well. They’re accurate and, as such, are being incorporated as credible signals in news and financial infrastructure. As of today, the Intercontinental Exchange (ICE) has invested $2B in Polymarket and now distributes Polymarket probability data through its Consolidated Feed; Kalshi's data is integrated into CNN's coverage so that reporters can better “tap into real-time prediction market data to better inform and fact-check [their] reporting,” and a Polymarket and Dow Jones’s partnership makes Polymarket's real-time data available across the Wall Street Journal, Barron's, MarketWatch, and Investor's Business Daily, with market updates integrating prediction market statistics. If prediction markets are increasingly treated as credible data, this makes them an inherently desirable channel to corrupt and spread misinformation through.
Take Russia's Internet Research Agency and its attempts to undermine the integrity of the 2016 U.S. electoral process. What made this operation effective wasn't in its ability to change votes directly but rather in its ability to generate content that looked like organic public sentiment (thousands of fake accounts on Facebook and Twitter posting divisive and polarized sentiments), which got picked up by the media as representative of real opinion and consequently shaped the informational environment that voters and decision-makers operated in. Prediction market prices sit in a similar signal category. They’re real-time, numerical, and now are increasingly treated as authoritative. A Polymarket probability listed alongside a Wall Street Journal article carries with it far more credibility than a viral tweet does.
The IRA’s Columbian chemicals hoax, from the first false tweet to confirmed debunking, took less than 24 hours, but in that time, thousands of tweets were generated and reposted, and public sentiment was already shifted. A large state player need not manipulate a market for long, but simply enough to support a flutter of tweets or a short-term misinformation campaign.
Looking forward, the risk lies most in the manipulation of prediction markets in parallel with traditional misinformation campaigns like the IRA’s. Such a campaign would require moving the price just enough, and at the right moment, to influence a media cycle or a decision-maker who reads the WSJ in the morning. Take the following hypothetical operation:
- A coordinated deployment of hundreds of anonymous wallets (likely via Polymarket as it requires no KYC), based in different geographic regions, well in advance of the planned operation. These wallets may sit inactive or trade periodically on various contracts, getting lost in the thousands of wallets created and traded on every day.
- When a bad actor decides to launch a misinformation campaign, they can simultaneously engage these various wallets to inject money into desired markets and shift a contract's probability by several points for, say, 48 to 72 hours, at a cost that is very well within the operational budget of a state-level influence operation.
- The price would then begin to inevitably self-correct as skilled traders arbitrage it back. However, the ability of these skilled traders to reverse the price and the time it would take them to do so could be greatly drawn out by the simultaneous misinformation campaign. While the 3.14% of skilled accounts undoubtedly use a mix of high-frequency mechanisms to drive information discovery, they still do so on public information. If false information successfully crowds out the truth, the time it would take for the price to revert to accurate levels would be significantly drawn out. In that time, whatever signal the bad actor was attempting to send had already been created and read.
- Major news outlets and large financial institutions reliant on ICE data now hold tampered information. A diplomat may walk into a room with a different sense of where American public sentiment stands on a policy he is about to negotiate. A defense contractor may accelerate procurement decisions after seeing ICE prediction market data that indicates firmer public belief of US forces entering Iran by a certain date.
The possibilities of how such operations may be carried out are practically limitless. But as long as the means to carry one out exist, so does the risk that someone will. I’m confident that such manipulation within prediction markets will become an attractive tool for bad actors to use in the near future, that is, if they are not already being utilized.
[^1]: Paraphrased from James Surowiecki’s The Wisdom of Crowds


