When AI learns to forge public opinion, how will the prediction market respond to the major test of manipulation?

Author: Andy Hall, Professor at Stanford Graduate School of Business and Hoover Institution

Translation: Felix, PANews (Content has been edited)

Imagine this scenario: It’s October 2028, and Wans and Mark Cuban are neck and neck in the presidential election. Wans’s support rate in the prediction market suddenly begins to soar. CNN, having partnered with Kalshi, provides round-the-clock coverage of the prediction market prices.

Meanwhile, no one knows the initial reason for the price surge. Democrats insist the market has been “manipulated.” They point out that a large volume of suspicious trades, without any new polls or other obvious reasons, has driven the market toward supporting Wans.

The New York Times also reports that traders backed by Saudi Arabia’s sovereign wealth fund have placed big bets on the election market to influence CNN’s favorable coverage of Wans. Republicans, on the other hand, argue that the prices are reasonable, noting there’s no evidence that the surge will affect the election outcome, and accuse Democrats of trying to suppress free speech and censor truthful information about the election. The truth remains uncertain.

This article will explain why such scenarios are highly likely to occur in the coming years—despite the rarity of successful prediction market manipulations and the almost complete lack of evidence that they influence voter behavior.

Attempting to manipulate these markets is inevitable, and when manipulation occurs, the political impact could far exceed the direct effect on election results. In an environment where any anomaly is viewed as a conspiracy, even brief distortions can trigger accusations of foreign interference, corruption, or elite collusion. Panic, accusations, and loss of trust may overshadow the actual impact of the initial actions.

However, abandoning prediction markets would be a mistake. As traditional polls become more fragile in an AI-saturated environment—characterized by extremely low response rates and pollsters struggling to distinguish AI responses from genuine human respondents—prediction markets offer a useful complementary signal. They aggregate dispersed information and are driven by real financial incentives.

The challenge lies in governance: building a system that preserves the informational value of prediction markets while reducing abuse. This may involve ensuring broadcasters focus on reporting markets that are harder to manipulate and more active, encouraging platforms to monitor for signs of collusion, and shifting how market fluctuations are interpreted—approaching markets with humility rather than panic. If achieved, prediction markets can evolve into a more robust and transparent component of the political information ecosystem: helping the public understand elections rather than fueling distrust.

Learning from history: Be wary of attempts to manipulate markets

“Now everyone is watching the betting markets. Their fluctuations are of intense interest to ordinary voters, who cannot directly gauge public sentiment and can only blindly rely on the opinions of those who bet hundreds of thousands of dollars in each election.” — The Washington Post, November 5, 1905.

In the 1916 presidential election, Charles Evans Hughes led Woodrow Wilson in the betting markets in New York. Notably, during that era in American politics, media frequently reported on betting markets. These reports cast a long shadow of potential manipulation. In 1916, Democrats did not want to be seen as behind, claiming the betting markets “were manipulated,” and the media covered these claims.

The threat of election manipulation has never disappeared. On the morning of October 23, 2012, during Barack Obama’s and Mitt Romney’s campaigns, a trader placed a large order on InTrade to buy Romney shares, causing his price to surge about 8 points—from just below 41 cents to nearly 49 cents—implying an almost tied race if one believed the price. But the price quickly retreated, and the media paid little attention. The identity of the manipulator was never confirmed.

Sometimes, you even see individuals openly explaining their logic for attempting to manipulate the market. A 2004 study documented a deliberate market manipulation case during the 1999 Berlin state election. The authors cited a real email from local party officials urging members to bet on the prediction market:

“‘Daily Mirror’ (one of Germany’s largest newspapers) publishes a political stock market (PSM) daily, with the current FDP (Free Democratic Party) trading price at 4.23%. You can view the PSM online via http://berlin.wahlstreet.de. Many citizens do not see PSM as a game but as a poll result. Therefore, it’s important that the FDP’s price rises in the final days. Like any exchange, the price depends on demand. Please participate in the PSM and buy FDP contracts. Ultimately, we all believe in our party’s success.”

These concerns re-emerged in 2024. On the eve of the election, The Wall Street Journal published an article questioning whether Trump’s advantage on Polymarket—seemingly far exceeding his polling support—was the result of improper influence: “Large bets on Trump may not be malicious. Some observers think it’s just a big gambler who believes Trump will win and wants to make a big profit. However, others see these bets as an influence operation aimed at generating buzz for the former president on social media.”

The scrutiny in 2024 is particularly intriguing because it raises fears of foreign influence. The results indicated that a French investor placed bets that pushed up Polymarket prices—though, despite speculation, there was little reason to believe this was manipulation. In fact, the investor conducted private polls and seemed focused on making money rather than manipulating the market.

This history reveals two themes. First, cyberattacks are common and likely to occur again in the future. Second, even if attacks are ineffective, some can still incite fear and panic.

How much impact do these attacks have?

Whether these measures influence voter behavior depends on two factors: whether manipulation can effectively impact market prices, and whether changes in market prices influence voters’ actions.

Let’s explore why market manipulation (if possible) could help achieve political goals: because it’s not as obvious as people think.

Here are two ways prediction markets might influence election outcomes.

Herd Effect

The herd effect refers to voters tending to support candidates who appear to be winning, whether due to conformity, the satisfaction of supporting the winner, or believing that market odds reflect candidate quality.

If popularity helps a candidate gain more support, then reporting prediction market prices in the news creates an incentive to push those prices higher. Manipulators might try to inflate the perceived chances of their favored candidate, hoping to trigger a feedback loop: market price rises → voters perceive momentum → voters shift support → price rises again.

In the Wans-Cuban example, the manipulator’s goal was to make Wans look stronger, which would help him win.

Satisfaction Effect

On the other hand, if voters support a candidate who is far ahead, they might choose not to vote. But if the race is close or their preferred candidate seems likely to lose, they may be more motivated to vote. In this case, widespread prediction market signals can create a market pressure to keep the support level near a 50-50 split. Once the market begins favoring a candidate, traders will see that supporters of that candidate are losing enthusiasm, which can pull the price down.

This also facilitates market manipulation. Leading candidates worried about over-optimistic supporters might quietly buy their opponents’ shares to tighten the market and suggest a more competitive race. Conversely, supporters of trailing candidates might further depress their shares to make the opposition appear more likely to win, discouraging supporters of the other side from voting. In this way, the market becomes a self-fulfilling prophecy: the signals meant to reflect expectations instead undermine them.

Despite considerable controversy, some argue Brexit is an example of this phenomenon. As a report from the London School of Economics notes: “It is well known that polls influence turnout and voting behavior, especially when one side appears to have a commanding lead. It seems that more supporters of Remain chose not to vote, perhaps because they believed Remain would win.”

Voters are not very concerned about the closeness of elections

But the issue is that, even if herd or satisfaction effects exist, existing evidence suggests their influence is usually small. U.S. elections are quite stable—mainly driven by party loyalty and fundamental factors like the economy—so if voters reacted strongly to claims about who is leading, election results would be more chaotic. Moreover, when researchers try to directly alter perceptions of election closeness or importance, behavioral effects are always limited.

For example, studies by Enos and Fowler on a state legislative race in Massachusetts, which ended in a tie, found that informing some voters that the last election was decided by just one vote had minimal impact on turnout. Similarly, Guber et al. showed in large-scale field experiments that presenting different poll results changed perceptions of competitiveness but had little effect on turnout. A study of Swiss referendums found a slightly larger effect: highly publicized close polls could marginally increase turnout—by only a few percentage points.

It’s possible that, at times, signals of a close race do influence some voters to change their minds, but this effect is likely very small. This does not mean election fraud should be ignored, but rather that attention should be paid to subtle influences in tightly contested races, rather than distortions that turn evenly matched elections into landslides.

Manipulating markets is difficult and costly

This leads to the second question: How hard is it to manipulate prediction market prices?

Research by Rhode and Strumpf on the Iowa electronic market during the 2000 election found that attempts to manipulate are costly and difficult to sustain. In a typical case, a trader repeatedly placed large buy orders to push the price toward their preferred candidate. Each attempt temporarily changed the odds, but was quickly exploited by arbitrageurs who restored the price to normal levels. Manipulators invested heavily but suffered losses, while the market showed strong mean reversion and resilience.

This is crucial in the hypothetical Wans-Cuban scenario. Manipulating the presidential market in October would require substantial funds, and many traders would be ready to sell once prices surged. Such small fluctuations might persist until CNN broadcasts the story, but by the time CNN anchor Anderson Cooper discusses it, prices could have already fallen back to initial levels.

However, when market liquidity is low, the situation changes. Researchers have shown that in low-liquidity environments, long-term prices can be manipulated: no one can prevent such manipulation.

Recommendations for dealing with manipulation

While there is some evidence that attempts to manipulate major election markets are unlikely to have significant effects, this does not mean we can be complacent. In a world where prediction markets are integrated with social media and cable news, the impact of price manipulation could be greater than ever. Even if the direct influence is limited, such concerns can affect the collective perception of the fairness of the political system. How should this be addressed?

For broadcasters:

Implement liquidity thresholds. CNN and other news outlets should focus on reporting prices from active markets, as these are more likely to reflect accurate expectations and are harder to manipulate; avoid reporting prices from illiquid markets, which are less accurate and easier to influence.

Incorporate other election signals. News organizations should also monitor polls and other indicators of electoral expectations. Although these have their own flaws, they are less susceptible to strategic manipulation. When significant discrepancies between market prices and other signals are detected, media should investigate potential manipulation.

For prediction markets:

Develop monitoring capabilities. Establish systems and personnel capable of detecting deceptive trades, false trades, sudden surges in unilateral trading volume, and coordinated account activity. Companies like Kalshi and Polymarket may already have some of these capabilities, but if they aim to be responsible platforms, they should invest more resources.

Intervene during abnormal price swings. This includes implementing simple circuit breakers in illiquid markets to handle sudden price changes, and temporarily halting trading and conducting auction reconsolidation when prices behave abnormally.

Improve price indicator robustness. Use time-weighted or volume-weighted prices for public display to reduce susceptibility to manipulation.

Enhance transparency continuously. Transparency is vital: publish metrics on liquidity, concentration, and abnormal trading patterns (without revealing personal data), so journalists and the public can discern whether price movements reflect genuine information or order book noise. Large markets like Kalshi and Polymarket already show order books, but more detailed metrics and user-friendly dashboards would be very helpful.

For policymakers:

Combat market manipulation. The first step is to clarify that any attempt to manipulate election prediction markets—aimed at influencing public opinion or media coverage—falls under existing anti-manipulation laws. When unexplained large price swings occur before elections, regulators should act swiftly.

Regulate foreign and domestic political influence. Given the susceptibility of election markets to foreign interference and campaign finance issues, policymakers should consider two safeguards:

(1) Monitor foreign manipulation by tracking traders’ nationalities, leveraging existing U.S. KYC laws that are crucial for prediction market operations.

(2) Establish disclosure rules or bans for expenditures related to campaigns, PACs, and senior political figures. If manipulation involves unreported political spending, regulators should treat it as political expenditure.

Conclusion

Prediction markets can make elections clearer rather than more chaotic—provided they are established responsibly. The partnership between CNN and Kalshi signals that future market signals will become part of the political information environment alongside polls, models, and reporting. This is a real opportunity: in an AI-saturated world, tools are needed to extract dispersed information without distortion. But this future depends on good governance, including liquidity standards, regulation, transparency, and more cautious interpretation of market dynamics. If these aspects are managed well, prediction markets can improve public understanding of elections and support a healthier democratic ecosystem in the age of algorithms.

Related reading: A decade of prediction markets—who will be the next?