New research shows that large AI models often fail to replicate human trading efficiency in complex markets. For investors, this raises questions about the effectiveness of AI in automated trading and the potential operational risks for financial firms.
In traditional trading setups known as double auctions—where buyers and sellers meet to reach a fair price—large artificial intelligence models are finding it difficult to match human efficiency. While human traders in historical studies often arrive at competitive price levels quickly, recent tests show that sophisticated large language models (LLMs) behave quite differently. Instead of naturally steering the market toward a stable price, these large AI models often prioritize securing tiny, marginal extra profits over completing transactions. This tendency causes trading volumes to drop and prevents the market from functioning as smoothly as it does with human participants.
Interestingly, the research found that smaller AI models were often more adaptable and effective at closing gaps between buy and sell prices than their larger, more powerful counterparts. This creates a paradox for financial institutions: simply deploying the most complex or expensive AI technology does not necessarily guarantee better market performance or faster trade execution.
For investors, this finding highlights a critical challenge for financial firms that are investing heavily in automation. A 2026 report from McKinsey found that despite the widespread use of AI, only 37% of companies are seeing a positive impact on their earnings (EBIT). This suggests that many organizations are struggling to convert AI spending into actual financial returns. When AI systems are used for high-stakes tasks like automated pricing or negotiation, inefficient decision-making by these models can lead to poor execution and, in extreme cases, market instability.
Regulators are already paying close attention to these risks. The Bank of England’s Financial Stability Board has raised concerns about the use of frontier AI in financial systems, specifically citing the danger of 'black box' decision-making—where the logic behind a machine's trade is unclear—and the potential for these systems to trigger disorderly market corrections. In India, both the National Stock Exchange (NSE) and the Bombay Stock Exchange (BSE) are actively developing oversight mechanisms to detect risks like potential market manipulation and misinformation. The NSE, for instance, reported technology spending of ₹1,315 crore in FY26, largely directed toward infrastructure and capacity expansion to handle these modern market challenges.
As financial firms continue to integrate AI into their trading desks, investors should look beyond just the amount of money spent on technology. The key monitorable will be how these institutions manage the operational risks associated with automated systems. The inability of some AI models to handle basic market mechanics suggests that human oversight and robust governance will remain essential, even as automation becomes more common. The industry's ability to refine these models and align them with stable market practices will be a major factor in determining whether AI truly becomes a long-term efficiency booster for the financial sector.