When Wall Street strategists rely on the same set of AI: trades become more crowded, risks more concentrated, and mistakes spread faster.

Wall Street's accelerated adoption of artificial intelligence (AI) is reshaping market microstructure, sparking a new round of industry concerns about increasingly crowded trading, systems susceptible to deception, and uncontrolled risk exposures.

As hedge funds and wealth management firms rush to use similar AI models and datasets to gain an investment edge, the divergence in market participants' views is narrowing. Recent research shows that this trend toward algorithmic homogenization is leading to highly converged portfolios and sharply shortening the lifespan of profitable trading signals, directly threatening active fund managers' ability to earn excess returns.

Meanwhile, AI-driven trading systems have revealed significant vulnerabilities and blind spots. Multiple tests show these models are not only prone to making poor judgments when confronted with subtly manipulated financial information, resulting in severe single-day net asset drawdowns, but also systematically bear risk volatility far beyond expectations.

These findings mark a shift in the focus of discussions about AI in the financial industry. The central concern has moved from "can technology help investors beat the market" to "how will market structure evolve when many investors rely on the same machine."

Strategy Homogenization Shortens Profit Cycles

The effectiveness of financial markets is built upon investors' disagreements, but wide adoption of AI is breaking this premise.

New York University researchers Shuchen Meng and Xupeng Chen analyzed nearly one million institutional fund holdings and found that as AI becomes prevalent in the investment industry, portfolio similarity continues to rise, especially among institutions that heavily use the technology.

This homogenization directly affects market structure. Their research model indicates that before widespread AI adoption, a profitable trading signal could last five to seven years, but now its excess return halves in about 18 months. As more investors reach the same conclusions almost simultaneously, today’s profitable strategies quickly turn into tomorrow’s crowded trades.

"Every additional edge AI participant incrementally shortens the lifespan of all exploitable patterns," Meng and Chen point out in their paper, "When everyone uses similar AI, collective results will differ in nature from the sum of individual interests."

Buy-side institutions' reliance on AI continues to deepen. A survey by the Alternative Investment Management Association (AIMA) last year showed that 58% of fund managers expect to rely more on AI in portfolio construction, compared with just 20% two years ago.

Information Manipulation Exposes System Vulnerabilities

In addition to exacerbating crowded trades, AI models’ reliance on input data also creates new single-point-of-failure risks. University of Liechtenstein researchers Advije Rizvani, Giovanni Apruzzese, and Pavel Laskov designed ten trading models based on large language models (LLMs) that use sentiment analysis to predict stock prices.

Although these models all delivered positive returns over a 14-month investment period, they proved completely defenseless against manipulated information.

Researchers made subtle changes to financial news headlines that were almost undetectable to human readers, such as swapping visually similar letters or embedding hidden text, successfully tricking all models. In the worst case of manipulating a single stock on a single day, the overall model return dropped by about 18 percentage points.

"A wrong decision can spread to other days and affect other decisions the system is making," said Rizvani. "Even if just for one day, it could lead to very catastrophic consequences."

Uncontrolled Risk Exposure and Overconfidence

While inheriting human traders’ analytical abilities, AI also seems to inherit humanity’s oldest weakness: overexposure to risk.

Jerry Bell, Victor Haghani, and James White of Elm Partners Management tested four popular AI models in a simulated trading challenge. After reading the front page of The Wall Street Journal, Claude and ChatGPT achieved an accuracy rate over 50% in predicting S&P 500 and US Treasury market directions, matching top macro traders.

However, the experiment revealed a crucial flaw. All four models consistently took too much risk, with daily return volatility reaching 20% to 40%, well above the recommended risk range of 7% to 15% set for them.

"We train AI to act like humans, and then we find AI really is like humans—overconfident, building excessive positions," Haghani commented.

As financial AI research moves from "can machines compete with humans" to "how investors compete through the same machines," market participants need to reassess the real costs brought by technology. As Apruzzese warned, "Blindly trusting large language models to make wise decisions is unwise. If everyone adopts AI simply because they think it's good or can help them make money, without considering the consequences, they may face enormous potential losses."

Risk Disclaimer and TermsThe market carries risks; investment requires caution. This article does not constitute personal investment advice, nor does it consider individual users’ specific investment goals, financial situation, or needs. Users should determine whether any opinions, viewpoints, or conclusions in this article fit their individual circumstances. Investing based on this content is at your own risk.