FRED data and AI agents: Waller on half of visits being bots

The news

Federal Reserve Governor Christopher Waller told FRED Con 2026 in St. Louis on October 1 that AI agents now generate about half of visits to FRED, the Federal Reserve Economic Data platform, and that traffic is growing 150 percent a year, mostly from AI and bots. He announced that users seeking certain data published by the Federal Reserve Board will soon be routed to FRED and its tools, and described a new connector for AI systems.

Key takeaways:

• FRED hosts more than 850,000 data series from governments and other sources and reached 18 million unique users in 2025.
• About half of FRED visits come from AI agents retrieving data; the other half are human.
• The St. Louis Fed has introduced the FRED MCP Connector, built on the Model Context Protocol, so users can retrieve FRED data with an AI agent of their choice.
• Waller warned that AI can hallucinate, introduce bias and misattribute sources, while noting that agents have generally been accurate describing charts and explaining concepts.

Our analysis

Waller, who spent part of his career at the St. Louis Fed, called FRED the greatest public good the Federal Reserve has created. He traced its history: a typed weekly digest mailed starting in 1961 by research director Homer Jones, then telephone updates, a computer bulletin board in 1991 and a website in 1995 with 865 series and about 6,000 users a week. The point of the history is that FRED has repeatedly been reshaped by technology, and he places AI in that line.

What differs now is the direction of the risk. Earlier advances mainly helped distribute data. AI can also create content, so it can state conclusions with confidence while hiding assumptions, citing sources wrongly or confusing correlation with causation. FRED’s managers do not know whether agents correctly credit the original data producer or FRED, and testing has sometimes shown credit going to FRED rather than the source. For a platform whose value is accurate reproduction of data from other organizations, attribution is part of the product.

The traffic figures make the problem concrete. Growth of 150 percent a year would more than double the load annually, and the St. Louis Fed has to design for a viewer that reads a chart very differently from a person. That is the reason for the connector: it lets users bring their own AI, as the FRED team puts it, to find and retrieve data, while FRED adapts presentation so that agents receive accurate, well-documented data. The team is also developing skills to help agents work with economic data and linking FRED to FRASER, the archive of Federal Reserve material, and to ALFRED, which stores data as first reported and in earlier forms.

The announcement about the Board’s data matters for people who rely on Fed releases. Waller said users looking for certain Board-published data will be routed to FRED and its tools. He did not list which datasets, or when.

Room for disagreement

Routing Board data through FRED concentrates the distribution of official statistics on one platform. Waller argues this is a strength, because FRED offers a common, consistent channel without commercial interest and guarantees faithful reproduction. A skeptic could see single-point dependence as a risk, especially while the traffic load is rising so fast.

He is also candid that AI errors are a live problem. The same agents that describe charts well have misattributed data in testing. Which effect dominates as use grows is not yet known, and the speech offers no accuracy rates for agent-generated answers.

What to watch

• Which Board datasets are routed to FRED and the timing of the change.
• Demonstrations and documentation for the FRED MCP Connector, shown later at the conference.
• Any measures FRED takes on attribution or on managing AI traffic.
• Progress on links between FRED, ALFRED and FRASER for AI queries.