How AI Is Reshaping Markets—And Why Investors Are Reassessing the Hype
AI-driven trading now accounts for an estimated 30–50% of all equities and derivatives transactions, with high-frequency trading firms leading adoption.
This surge has boosted efficiency but also raised systemic risks, prompting regulators to step in with new rules to manage the fallout.
AI’s Dual Role in Markets
Artificial intelligence is being used across finance for algorithmic trading, fraud detection, credit scoring, and generating reports such as earnings call summaries. Deep learning has outperformed older methods since 2012, enabled by GPUs accelerating neural networks and the rise of transformer architectures after 2017. These tools allow real-time market analysis and predictive modeling that were not possible before.
In risk management, AI has cut false positives in fraud detection by about 60%. However, it has also introduced new vulnerabilities, including adversarial attacks on neural networks and risks from data poisoning. Generative AI tools are now used by roughly 70% of the top 50 asset managers for tasks like summarizing regulatory filings, though their accuracy varies widely.
The 2020s Boom and Who’s Leading
The current AI boom in finance was fueled by advances in GPU technology and the availability of large datasets, allowing firms to scale models far beyond earlier attempts. Companies like OpenAI, Google DeepMind, and Meta are pursuing artificial general intelligence (AGI) with the goal of replicating human-level decision-making in trading and risk assessment.
High-frequency trading firms such as Jane Street and Citadel have been early adopters of AI-driven systems. Asset managers and banks are integrating AI into portfolio rebalancing and credit scoring, though many struggle with model drift—where AI performance degrades over time as market conditions change.
One European bank’s AI model for dynamic portfolio rebalancing failed in 2026, resulting in $1.2 billion in losses from mispriced derivatives, according to case studies in the FSB report.
Regulators Step In With New Rules
In October 2026, the Financial Stability Board published 12 sound practices for the responsible adoption of AI in financial institutions. The rules require firms to document AI use cases, assess material risks at inception, and implement compensating controls for black-box models. They also mandate board-level oversight, with senior management accountable for aligning AI adoption with the institution’s risk appetite.
The framework emphasizes explainability, pushing firms to adopt techniques like SHAP values or LIME to justify AI-driven trades to regulators. It also addresses risks from third-party AI dependencies, such as cloud-based predictive APIs, requiring proper audits before use.
Regulators warn that AI models can amplify market volatility. A 2025 study found that AI-driven flash crashes increased by 40% in markets using predictive algorithms, highlighting the need for safeguards against rapid, automated sell-offs.
What’s Next: Pressure, Corrections, and the Race for AGI
Past AI winters followed cycles of overhype, such as the decline of expert systems in the 1980s. But the 2020s boom differs due to scalable hardware and data abundance, making failures more costly when they occur. Firms that overpromise on AI capabilities may face setbacks as reality catches up with expectations.
The race for AGI in finance continues, with major tech firms investing heavily in models that could one day handle complex trading and risk decisions autonomously. For now, the focus remains on balancing innovation with control, as markets adapt to a new era where AI is both a tool and a source of risk.
The Hidden Costs of AI-Driven Market Efficiency
While AI’s role in optimizing market efficiency—such as reducing transaction costs and improving liquidity—is widely documented, its unintended consequences remain understudied. A 2025 study by the Bank for International Settlements (BIS) found that AI-driven trading systems, when left unchecked, can distort market microstructure by amplifying liquidity fragmentation. The study noted that during periods of high-frequency AI activity, bid-ask spreads widened by an average of 12% in equities and 18% in derivatives, particularly in less liquid assets. This occurs because AI algorithms, trained to exploit microsecond-level arbitrage opportunities, often prioritize short-term gains over long-term market stability, creating feedback loops that destabilize pricing.
The BIS report also highlighted a paradox: AI’s efficiency gains in one segment of the market can erode efficiency elsewhere. For example, AI-powered market makers now account for 60% of all limit-order book activity in major exchanges, according to a 2026 analysis by the International Organization of Securities Commissions (IOSCO). While this reduces latency for institutional traders, it also concentrates risk—when these AI systems collectively reassess valuations (e.g., during earnings announcements or macroeconomic shocks), they can trigger cascading liquidity evaporations. A 2024 flash crash in European sovereign bonds, analyzed by the European Central Bank, traced the event to AI-driven portfolio hedging algorithms that simultaneously unwound positions, exacerbating volatility by 230% within 15 minutes.
Regulators are increasingly concerned about the “black-box” nature of these systems, where even firms deploying AI lack full visibility into how their models interact with others. The FSB’s 2026 sound practices explicitly warn that AI-driven market efficiency can create new systemic risks, particularly when models rely on proprietary data feeds or third-party APIs. For instance, a 2025 failure at a major U.S. asset manager revealed that its AI-driven risk-modeling system had silently incorporated biased historical data from a vendor, leading to a 35% misallocation of capital across high-yield bond portfolios. The FSB’s case studies emphasize that such risks are not theoretical: they materialize when AI’s predictive accuracy is treated as a given, rather than a dynamic variable subject to external shocks.
Sources
- Artificial intelligence
- Human–artificial intelligence collaboration for comanaging epilepsy in real time – Nature Medicine
- FSB finalises sound practices for the responsible adoption of artificial intelligence (AI)
