AI Decision-Making Risks Highlighted by False Tariff News

On April 7, 2025, at 10:08 a.m. Eastern Time, U.S. stocks experienced a significant downturn. A false claim began circulating on the social media platform X, suggesting that Kevin Hassett, director of the White House National Economic Council, had stated President Donald Trump was considering a 90-day pause on tariffs for all countries except China. Market-focused news outlets rapidly amplified this unconfirmed report. CNBC repeated the claim on air, and Reuters subsequently published a report citing CNBC's broadcast. This news led to an immediate market rally, with the S&P 500 briefly recovering its losses as traders reacted to the perceived de-escalation of the trade war.

However, the report was inaccurate. Kevin Hassett had not made the statement attributed to him. Following a White House statement labeling the report as "fake news," CNBC issued a correction, and Reuters retracted its headline. This reversal sent the market in the opposite direction. According to Dow Jones Market Data, the U.S. stock market experienced a swing of approximately $2.4 trillion between 10:08 a.m. and 10:18 a.m. on April 7, 2025. This incident occurred before the current widespread adoption of AI agents in financial decision-making but serves as a stark preview of potential future problems.

Wall Street has long utilized machine-readable news, enabling software to process information and execute trades almost instantaneously. The evolving landscape involves AI agents being tasked with more complex responsibilities before executing trades. These agents are capable of gathering information from diverse sources, interpreting findings, evaluating conflicting signals, and, depending on their granted authority, recommending or directly implementing actions. This capability elevates the challenge beyond merely ensuring the quality of information to also assessing its credibility, true meaning, relevance to a given task, and sufficiency of evidence for action.

Nathaniel Bradley, CEO of Datavault AI, emphasizes that "provenance isn't truth," a concept clearly illustrated by the April 7th episode. A system receiving the Reuters alert could identify Reuters as the publisher and CNBC as the source of amplification. However, the critical challenge for AI agents lies in discerning the veracity and implications of such information, especially when rapid, high-stakes decisions are involved. The $2.4 trillion market swing underscores the profound financial implications of AI acting on unverified or misinterpreted data, highlighting the need for robust validation and judgment mechanisms within AI trading systems.