Predicting the economic impact of artificial intelligence

In May 2026, Federal Reserve Board Member Lisa D Cook delivered a keynote speech at the Stanford Institute for Economic Policy Research, systematically expounding on the dual impact of artificial intelligence (AI) on the U.S. economy and financial system. The speech combined the current macroeconomic situation, focusing on artificial intelligence, explaining the transmission effects of the AI investment boom on inflation, employment, and growth. It analyzed the innovative value in the financial sector and various potential risks, while also introducing practical case studies of the Fed’s implementation of AI-based regulatory analysis. With an optimistic yet prudent attitude, Cook proposed a development approach that advances innovation and risk control in parallel. The content offers both a macro perspective and practical industry references. The Institute of Financial Technology at Renmin University of China compiled the core research sections.

Macroeconomic Situation and the Transmission Effects of Artificial Intelligence

In her speech, Lisa first tied the discussion to the current trajectory of the U.S. economy, analyzed the multiple impacts brought by AI around the Fed’s dual policy goals (dual mandate), and used this as a basis to interpret the logic behind the implementation of current monetary policy.

(1) Inflation: Short-Term Shocks and Persistent Pressure

The U.S. inflation problem is prominent. As of the twelve months to April 2026, the Personal Consumption Expenditures (PCE, Personal Consumption Expenditures price index) increased by 3.8% year over year, which is clearly higher than the 2% inflation target set by the Fed. After excluding food and energy categories with relatively high volatility, the year-over-year increase in the core Personal Consumption Expenditures price index (core PCE) reached 3.3%, setting a new high since 2023. The direct trigger for this round of upward inflation was the rise in refined oil prices caused by the Iran situation. From a theoretical perspective, tariff disputes and geopolitical shocks are considered short-term disturbances; market expectations in the crude oil futures market also generally believe that international oil prices will decline toward the end of this year.

In the speech, Lisa emphasized that short-term price shocks still carry the risk of evolving into medium- and long-term inflation. Firms may easily incorporate phase-specific price increases into a normalized pricing framework, and workers may refer to current price levels during wage negotiations, thereby giving rise to wage-price spiral risks. At the same time, the AI investment boom on a large scale further exacerbates supply-demand imbalances in the market. The total amount of global enterprises’ data center construction plans has already exceeded $1.5 trillion, while most projects are still in the preparation stage. The continuous release of investment demand keeps pushing up the prices of upstream products such as chips and high-end hardware and software. Over the past year, wages for specialized construction trades have risen noticeably, and electricity and water prices have also increased by about 5% each. Beyond data centers, investments in fixed assets related to AI—such as robotics—will continue to expand, and medium- to long-term demand pressures will continue to support prices.

**

(2) Labor Market: Surface-Level Stability and Deep Risks

**

The labor market is currently generally stable. In April 2026, the U.S. unemployment rate was 4.3%. Since last summer, it has basically shown no major fluctuations. This figure matches the natural rate, implying that labor supply and demand are broadly in balance. Although there have been frequent reports of corporate layoffs, the number of initial unemployment insurance claims has remained low, and employment fundamentals are temporarily stable. However, Lisa pointed out that downside risks in the labor market are gradually accumulating. Economic uncertainty arising from Middle East geopolitical conflicts will suppress overall market demand. In response, companies may slow down hiring out of a wait-and-see mindset, causing the labor market at present to show characteristics of weak hiring intent. From a long-term development perspective, artificial intelligence may trigger the largest restructuring of employment in scale in decades. Although there has not yet been large-scale unemployment, the pace at which AI reduces jobs is likely—by a large margin—to be faster than the pace at which new jobs are created. Increased labor mobility will become an inevitable trend. Based on the Fed’s 2025 survey on small business credit, most small and micro enterprises have not yet changed their labor costs due to AI, but many companies already anticipate that firms will ultimately rely on AI to completely adjust operating models, meaning that changes in employment structure are only a matter of time.

**

(3) Growth: Optimism on Productivity and the Knowledge-Economy Dividend

**

On the dimension of economic growth, Lisa is optimistic. Over the past year, U.S. gross domestic product (GDP, Gross Domestic Product) growth has been strong, and labor productivity has surpassed the average level before the pandemic. Market entrepreneurial vitality has remained at a consistently high level. She combined this with endogenous growth theory and argued that artificial intelligence is a general-purpose technology with epoch-making value. Since World War II, the world’s long-term investment in the knowledge economy is now achieving a concentrated burst of innovative results with the help of AI. Once AI is systematically integrated into companies’ production processes, labor productivity will receive further reinforcement, providing strong support for steady growth in the U.S. economy in the near and medium term.

**

(4) Monetary Policy: Waiting in Stable Mode, Leaning Toward Tightening

**

Based on a comprehensive assessment of the macroeconomy, Lisa explained the core thinking behind monetary policy at this stage. From the perspective of risk management, the best choice is to keep the benchmark interest rate unchanged. The risk structure in the current economic operation is not balanced; upside inflation risks remain the main contradiction. According to the baseline forecast, inflation will gradually decline over the coming months, and the labor market will remain stable, so there is no need to adjust interest rates for now. However, the U.S. has been in a state where inflation has remained above the target range for five consecutive years. Once elevated prices become embedded in pricing and wage-setting mechanisms, it will create long-term hidden risks. Lisa stated clearly that if inflation falls short of expectations, the Fed will take rate-hiking measures; if the labor market shows a clear deterioration, it will also lower rates in a timely manner. Policy adjustments will fully rely on economic data to change flexibly.

Opportunities for Financial System Development Brought by Artificial Intelligence

When discussing AI’s value to the financial system, Lisa said that AI can comprehensively improve production efficiency, accelerate the iteration speed of innovation outcomes, and help emerging companies emerge and create new jobs—thereby easing inflation pressure from a macro perspective. Internally, the Fed places great emphasis on experimental exploration and innovative models. By drawing on Silicon Valley’s development experience, it takes the lead in building a research network for emerging technologies, sharing AI research results and implementation experience across the entire Fed system, and encouraging staff to explore new directions for applying artificial intelligence. It also tolerates trial-and-error during the innovation process. At this stage, the financial industry has already been among the first to apply AI to labor-intensive, resource-consuming traditional business segments. Scenarios such as compliance review, customer call centers, and back-office operations have all undergone intelligent upgrades, and work efficiency has improved significantly. AI tools also make data analysis more efficient and flexible. With intelligent coding technologies, financial institutions are able to address longstanding problems such as updating legacy code systems and integrating multiple systems. Large technology firms and financial institutions also use AI proactively to identify various cybersecurity vulnerabilities and strengthen system defense barriers. Looking ahead, the space for artificial intelligence to transform the financial industry is broad. By leveraging AI, it is possible to build customized financial products and design differentiated services according to different customers’ needs, so that complex financial products can reach more groups. For retail investors, AI analysis tools can help users capture market trends earlier and identify potential risks. As industry operating efficiency improves, more capital can flow into the credit and investment sectors, further activating the real economy and forming a virtuous economic cycle.

Financial Risks and Fragility Points Triggered by Artificial Intelligence

Lisa also objectively noted that technological innovation must come with various risks. If effective supervision and constraints are lacking, AI will amplify the weak links that already exist in the financial system, and it will also give rise to entirely new risks. AI-related financial risks mainly manifest in four areas.

The first type of risk is market risk caused by AI-driven algorithmic trading. Traditional algorithms rely on fixed code and simple rules to carry out high-frequency trading, and their operating models are relatively rigid. Generative AI and machine learning, however, have autonomous learning capabilities and can dynamically adjust trading strategies by combining historical data, real-time market information, and unstructured inputs such as text. Such new trading patterns can easily lead to convergence in trading behavior, creating an endogenous model collusion problem. This can lower the implementation threshold for market manipulation, further increase market concentration, disrupt normal market order, and threaten financial stability.

The second type of risk is the risk of industry transformation being transmitted to the credit market. AI technology has disrupted certain traditional industries, and this trend has already been reflected in the bond market. Driven by expectations of industry restructuring, the credit spreads of speculative-grade bonds in the technology sector continue to widen. After a leading AI company launched targeted products that impacted the software industry, market concerns about the related credit assets intensified, triggering large-scale redemptions. This caused clear disruptions to both on-exchange trading-style bonds and off-exchange perpetual bonds issued by business development companies.

The third type of risk comes from the debt leverage problem created by AI infrastructure construction. To roll out AI hardware facilities such as data centers, many technology companies have turned to debt markets for financing. Top technology companies frequently issue investment-grade bonds, while smaller data center operators rely on private credit and asset-backed markets to obtain funding. The borrowing scale in emerging technology fields continues to expand. Leverage ratios continue to rise, gradually accumulating systemic risks. Lisa added that although the overall leverage level has not yet reached the peak before global financial crises, the trend of continuously disorderly bond issuance still warrants high vigilance.

The fourth type of risk is cybersecurity risk, which is also the area attracting the most attention right now. The capabilities of large language models (LLMs, Large Language Models) and agentic AI are improving rapidly. Such technologies can be used to defend system vulnerabilities, but they could also be leveraged by malicious actors. Taking Anthropic’s Mythos model as an example: it can identify software vulnerabilities that were previously undiscoverable. If it is used by hackers, it would seriously threaten the security of financial institutions and critical infrastructure. At the same time, AI has significantly lowered the barrier to writing code, and the number of code segments grows quickly. This indirectly increases the burden on existing security review systems, and non-malicious cyber problems such as software malfunctions are also more likely to cause interruptions in financial services. Of course, AI is not only negative. Advanced intelligent tools can also build defense systems and repel cyberattacks. Overall, the industry is entering a game-like pattern in which offense and defense technologies upgrade in parallel.

AI Application Practices at the Federal Reserve

Lisa provided a detailed introduction to the Fed’s internal AI application practices. The Federal Open Market Committee (FOMC) currently does not use artificial intelligence to formulate monetary policy. However, across the Fed, AI has been widely applied to tasks such as financial stability monitoring and risk analysis. By developing and deploying intelligent tools, regulators can more precisely identify new types of risks derived from AI, and can also detect issues that are prone to be missed under traditional regulatory approaches.

The Fed’s AI practices are divided into two main parts. The first part is to set up specialized technical teams to study the opportunities and risks brought by technologies such as network security, AI, and quantum computing. The team uses large language models to conduct simulation experiments, exploring how generative AI affects investor behavior. The experimental results show that AI agents are more inclined to make judgments based on data and logic, which can effectively weaken blind follow-the-crowd behavior driven by animal spirits and reduce the probability of asset bubbles emerging. The research team also applies active knowledge distillation technology to develop lightweight AI models. Under the premise of keeping classification accuracy unchanged, it reduces computational costs by 80%, enabling efficient processing of massive volumes of unstructured text such as regulatory documents, financial reports, and news. In addition, staff members use Natural Language Processing technology to analyze the text of decades of The Beige Book, confirming that sentiment data in text can effectively predict economic recessions and significantly enhance macro risk early-warning capabilities.

The second part is to organize staff from the Board of Governors and 12 regional Federal Reserve Banks to conduct hands-on exercises with agentic AI, exploring the application of this technology in financial stability analysis. Agentic AI has the ability to reason autonomously, choose analytical methods, and independently complete complex tasks. In the identification of network-based risks, it performs far better than traditional manual analysis approaches. Constrained by limited manpower and computing resources, traditional methods find it difficult to comprehensively map complex financial network structures. In contrast, agentic AI can complete systematic scanning. Meanwhile, these tools can also batch set up and run various financial stability scenario simulations, completing work that previously required large amounts of human effort and time. To avoid errors in algorithmic judgment, the Fed builds multiple validation mechanisms into its system, requiring multiple AI agents to cross-argue and proactively incorporate opposing viewpoints, and then submitting the final results to researchers for review. This operating logic draws on the pattern of human thinking and academic research, balancing efficiency and accuracy.

Summary and Development Stance

Finally, Lisa said that the development of artificial intelligence must be grounded on three pillars: experimental exploration, governance rules, and risk control. At present, both financial institutions and technology companies are accelerating the deployment of AI technologies, while the pace of technological iteration continues to increase. Only if regulators personally practice and deeply understand the system’s operating logic can they precisely identify risks and guide the industry toward healthy development.

AI has strong analytical capabilities and can expand the boundaries of human work, but its advantages must be built on a sound governance framework. The best application model at this stage is human-AI collaboration: using AI to enhance human judgment, while embedding verification mechanisms in the system’s underlying layer to avoid algorithmic biases and decision-making errors. In this critical period of technological transformation, the Fed remains optimistic about the development of artificial intelligence, while also adhering to a prudent principle. On the basis of encouraging innovation and maintaining vitality, it firmly safeguards the bottom lines of macroeconomic stability and financial stability.