How AI Trading Bots Could Be Secretly Colluding, Raising Your Investment Costs

AI trading bots now execute many of the trades on Wall Street, but July 2025 research suggests these bots can unintentionally learn to work together in ways that drive up costs for everyday investors.

A July 2025 National Bureau of Economic Research (NBER) study found that even without human intervention, AI systems designed for trading and price-setting can independently develop collusive strategies, potentially shifting asset prices and reducing competition in the market. “They autonomously sustain collusive supra-competitive profits without agreement, communication, or intent,” the study’s authors, Winston Wei Dou, Itay Goldstein, and Yan Ji, wrote. “Such collusion undermines competition and market efficiency.”

The U.S. Securities and Exchange Commission and Congress have begun to recognize this risk and are pursuing new rules. In this article, we’ll explain how nonhuman entities can unwittingly collude in a market and how it might affect your portfolio.

Robots Gone Rogue

Building on previous studies, the NBER researchers used simulations to show that algorithms using reinforcement learning, a machine learning technique for optimizing strategies through repeated trial and error, learn to coordinate pricing, even as they’re not explicitly programmed to do so.

While the systems are sophisticated, the process itself isn’t. The study reveals that AI systems have two distinct types of collusive behavior, one the researchers dub “artificial intelligence” and the other they dub “artificial stupidity”:

• “Artificial intelligence” collusion: Algorithms learn to use price signals as a monitoring system. When one bot tries to undercut others, the rest automatically punish it by reverting to aggressive competition until the cheater falls back in line.
• “Artificial stupidity” collusion: The accidental kind, where learning biases cause bots to systematically avoid aggressive strategies. “Aggressive strategies, by their nature, are more exposed to noise trading shocks, making them especially vulnerable to this asymmetric learning dynamic,” the researchers explain.

While the researchers didn’t claim that this is happening now in the financial markets, what’s remarkable is how this mirrors classic cartel behavior. However, traditional cartels require meetings, agreements, and enforcement mechanisms. AI bots simply observe patterns and adapt.

A previous study by U.S. Federal Trade Commission researchers found something similar. The bots learned classic price-fixing tricks, like punishing competitors that tried to offer lower prices, then slowly raising prices back up once the “cheater” fell in line. This happened even when researchers made the trading situations more complicated and thus presumably harder to coordinate.

The Costs of AI Collusion

Since the study reveals that AI collusion can occur through “artificial stupidity,” this can make it almost impossible to distinguish from normal market behavior, since there’s no communication or explicit coordination to detect.

The consequences for everyday investors could be significant:

• Higher trading costs: Collusive behavior widens bid-ask spreads and increases trading costs for all market participants. Retail investors, who don’t have the sophisticated systems needed to detect subtle market manipulation, would likely face increased costs.
• Markets can be less efficient: When prices are artificially high, markets can’t do their main job: figuring out what things are worth.
• Broader economic effects: If bots converge on high-price strategies across major markets like commodities, housing, or equities, the outcomes could mirror historical cases of price-fixing with wide-reaching economic consequences.

What Can Retail Investors Do?

For everyday investors, AI trading collusion could directly impact your investment costs and returns. Should such AI collusion be found to have jumped into the real world of equities trading, here are some things you can do:

• Use limit orders: Unlike market orders, where the price can shift in market trading, limit orders set the price you’ll pay, thus reducing your exposure to potentially manipulated spreads.
• Focus on longer-term investing: Any collusion is far more likely to affect frequent traders. The less you trade, the less it would affect you.
• Diversify across asset classes and geographies: This would reduce your exposure to any single market where collusion might be more prevalent.

Bottom Line

AI-powered trading bots promise more efficient markets, but they carry the hidden risk of algorithmic collusion. This could unintentionally drive up trading costs and reduce competition for all investors.

While significant changes to regulations for further oversight of the market’s AI systems seem off the table in the mid-2020s, investors can take steps to protect themselves: use limit orders instead of market orders, focus on long-term investing rather than frequent trading, diversify across asset classes and geographies, and choose low-cost index funds to keep their trading frequency low.