A simulated Federal Reserve meeting using AI agents revealed that political pressure divided board members during discussions on interest rates, according to Reuters.
Researchers from George Washington University created virtual assistants modeled after real politicians and replicated the July 2025 meeting of the Federal Open Market Committee. They used historical positions of actual participants, their biographies, and public speeches.
The assistants were then tasked with processing current economic data and financial news to make decisions on the rate.
Under political pressure, the agents became polarized, and the number of dissenting opinions increased.
“This simulation shows that the Federal Reserve is only partially insulated from politics, and external factors can influence internal decisions, even in an institution operating under formal rules,” the study stated.
AI in Economic Processes
While few banks are ready to entrust monetary policy to AI systems, more are using the technology to optimize their operations:
- The Fed has conducted research using generative models to analyze FOMC meeting minutes;
- The European Central Bank employs machine learning to forecast inflation in the eurozone;
- The Bank of Japan uses AI for information gathering and in-depth economic analysis;
- The Reserve Bank of Australia is testing a new tool for drafting briefings on policy issues.
“It is important to understand: we do not use artificial intelligence to formulate or set monetary or other policies. Instead, we aim to leverage it to enhance efficiency and strengthen staff contributions in areas such as research and analysis,” said Michelle Bullock, head of the Reserve Bank of Australia.
Central banks see strategic value in AI and are actively experimenting with its application for data search and analysis, as noted in an April report by the Bank for International Settlements. Many are in the early stages of implementing the technology.
In Ukraine, the “world’s first” state AI agent was launched on the “Diia” portal. It not only answers questions but also provides services directly in the chat.