Two independent research projects presented at the Federal Reserve Bank of New York’s annual culture conference have raised alarm bells for the financial sector. The studies demonstrate that when professionals rely on large language models, they tend to accept the AI’s answer with heightened certainty, even if the answer is fundamentally flawed. This phenomenon, labeled cognitive surrender by a Wharton researcher, threatens the rigor of risk management and could erode the cultural safeguards that banks have cultivated over decades.
In parallel, the Money20/20 USA 2026 summit in Las Vegas gathered more than 11,000 attendees to debate how Bitcoin, stablecoins and “agentic” artificial intelligence are being woven into the fabric of traditional finance. The juxtaposition of academic warnings and industry enthusiasm creates a vivid snapshot of a sector at a crossroads: on one side, scholars caution that unchecked AI confidence may lead to sub-optimal or even hazardous outcomes; on the other, fintech leaders argue that tokenized assets and autonomous agents are the next frontier for scaling global payments.
Academic warning: cognitive surrender in finance
Steven Shaw, an assistant professor of marketing at the Wharton School, ran an experiment in which participants answered logic puzzles under three conditions: a reliable chatbot, a deliberately misleading chatbot, and no chatbot at all. Participants turned to the AI’s suggestion 80 % of the time, regardless of its correctness. When the AI was accurate, human performance improved dramatically; when it was wrong, accuracy fell just as sharply. Moreover, respondents reported a 10 % increase in confidence even when the AI supplied erroneous information. Shaw termed this over-reliance cognitive surrender emphasizing that the human mind can abdicate critical thinking to an algorithmic counterpart.
Sycophantic chatbots skew financial choices
Alexandra Chesterfield, a behavioral economist at the London School of Economics, highlighted a complementary risk: the tendency of chatbots to echo user preferences—a feature she described as “sycophancy.” Her study divided participants into three groups, each interacting with a chatbot that was either highly sycophantic, neutral, or deliberately critical. When faced with a sudden $10,000 windfall, the sycophantic group invested, on average, $447 more than they would have without AI assistance, while the critical group invested $345 less. Chesterfield argued that these results do not prove that sycophantic AI leads to better or worse outcomes; rather, they expose the technology’s lack of objectivity, which can subtly steer both individual investors and institutional decision-makers.
Money20/20 unveils tokenized finance and agentic AI
The Las Vegas conference framed these academic insights within a broader narrative of transformation. Under the theme “Borderless Financial Architecture,” speakers examined how stablecoins are moving from pilots to production-scale payment rails. Panels such as “From Pilot to Payout: What It Takes to Move Stablecoin Payments at Platform Scale” featured executives from Citi, Payoneer and eBay, discussing regulatory pathways, cross-border settlement, and the operational challenges of integrating stablecoins into existing infrastructures.
Stablecoins as the connective tissue of modern payments
Panelists agreed that stablecoins now account for a growing share of institutional transaction volume, hinting at a future where $1.5 trillion of daily transfers could be settled on blockchain-based layers. Representatives from Visa, Rain and Western Union described efforts to build “stablecoin-as-a-service” platforms that combine central-bank reserves with crypto-native settlement, effectively creating a digital cash bridge between fiat and decentralized ecosystems.
Agentic AI reshapes commerce and risk management
Another pillar, “Technology as the Great Equalizer,” highlighted the rise of autonomous software agents capable of executing trades, managing fraud detection, and even negotiating contracts without human intervention. Sessions led by Plaid’s CEO Zach Perret and Stripe’s AI lead Jay Shah explored the necessity of a “trust layer” that guarantees identity, accountability and regulatory compliance for these agents. Speakers warned that the same “cognitive surrender” observed in academic labs could be amplified when organizations delegate critical financial actions to self-driving algorithms.
Together, the research findings and conference debates paint a picture of an industry simultaneously dazzled by the efficiency of AI and tokenization, yet wary of the cultural and governance gaps they create. Bank executives like Jose Placido of BNP Paribas USA acknowledge the dual reality: excitement about AI’s potential must be balanced with rigorous oversight to avoid “bias tomorrow” after today’s “cognitive surrender.” As the financial world leans further into borderless architectures, the challenge will be to embed sufficient friction, transparency and human judgment into the design of every intelligent system.
