The Wall Street Quant Who Used AI to Predict Three Straight Market Downturns Is Sounding the Alarm Again

A 7,000-word document published on Substack started circulating early on February 23, 2026, somewhere in the glass-walled conference rooms and open-plan offices that line the streets close to the New York Stock Exchange. The majority of Wall Street traders were unaware of Citrini Research, the company that produced it. By the afternoon, shares of Uber, Mastercard, American Express, and Blackstone were all declining, and the Dow had fallen more than 800 points. The author of the document had referred to it as a scenario. Not a forecast. Apparently, the distinction did not reassure the market.

What Citrini Research had written was a meticulously crafted picture of a near future in which autonomous AI systems, or “agents” as the industry refers to them, systematically destroy the American economy from within. jobs. markets for credit. underwriting for mortgages. The reasoning was not that AI would fail, but rather that it would succeed so rapidly and thoroughly that the current economic structure would not have time to adapt. The article described it as a feedback loop without a brake. Throughout, the language was fictitious. The reaction of the market was not.

In a market that has been searching for reasons to sell, it would be simple to write this off as a single instance of volatility. However, that reading overlooks the story’s more intriguing aspect, which is its history. Before the February report, Citrini Research had been raising similar concerns for years, highlighting the potential for AI-driven market disruption at times when the Nasdaq was still rising and the general consensus was still optimistic. The company’s approach, which is based on deep learning systems that monitor worldwide datasets for what its researchers refer to as early indicators of “herd behavior” changes, had identified the circumstances leading up to three distinct market downturns with sufficient lead time to be significant. Three times. in order. That is either a truly potent analytical framework or a remarkable coincidence; eventually, the pattern takes precedence over the distinction.

The relationship between Wall Street and this type of research is complex. Hundreds of PhDs in mathematics and physics are employed by the big investment banks’ quantitative departments to create models that process market data faster than any human analyst. Independent researchers are unable to duplicate the infrastructure that Goldman Sachs, Morgan Stanley, and Citadel have for this kind of work. Nevertheless, calls from outside those buildings have consistently provided the correct direction. Not the institutions holding the mortgage-backed securities, but a small group of independent analysts made the most famous predictions about the 2008 mortgage crisis. In the weeks preceding the 2020 COVID crash, no major bank’s risk model identified it. Smaller, more focused research operations appear to handle institutional blind spots more skillfully, perhaps because they are not juggling the conflict between analytical integrity and client relationships.

Although the specific thesis of the February report—that AI agents could cause systemic disruption rather than gradual disruption—is not new in academic circles, it had never before been expressed in a way that made money. An AI-driven equity correction was already listed as one of three major risks for the year in Bloomberg’s 2026 investment outlook, along with a rise in inflation and geopolitical energy shocks from the Iran conflict. The market was already jittery about AI valuations, watching ServiceNow and other AI-related names trade at multiples that required flawless execution to justify, and taking in the news that Anthropic’s Mythos model had sparked new cybersecurity concerns among major banks when the Citrini piece arrived. In other words, it landed at precisely the wrong—or, depending on your position, the right—moment.

It was described by the Wall Street Journal as a thought experiment that alarmed investors who were “already wary of tech disruptions.” Although accurate, that framing is a little too cozy. A thought experiment that causes the Dow 800 to plummet in the afternoon is more than just a cerebral exercise. At the very least, it is a signal about what the market considers to be plausible, and plausible enough to set prices right away without waiting for the scenario to start playing out. Preemptive pricing of that type is a data point in and of itself. Selling off on hypotheticals is not what markets are meant to do. When they do, it typically indicates that the hypothetical addressed a concern that participants had previously expressed in private but had not voiced aloud.

Watching this unfold gives me the impression that the anxiety fueling the reaction to the Citrini report is less about the particular situation and more about a broader uneasiness that has been growing since DeepSeek’s rise last year raised the first significant concerns about competitive moats in AI. The market had been pricing AI as a one-way story: invest in infrastructure, wait for revenue, then repeat. However, the potential for the infrastructure to become destabilizing rather than merely generative adds a variable that is difficult to account for in any discounted cash flow model. It appears that the quant at Citrini, whoever created the initial model, anticipated that variable. The market is still debating how to handle it.