This semi-monthly column highlights news, government documents, NGO/IGO papers, conferences, industry white papers and reports, academic papers and speeches, and central bank actions on the subject of AI’s fast paced impact on the banking and finance sectors. The chronological links provided are to the primary sources, and as available, indicate links to alternate free versions.
NEWS
Where Global Economies Sit in the AI Stack – Franklin Templeton, March 3, 2026. Dina Ting, CFA, Head of Global Index Portfolio Management for Franklin Templeton ETFs, writes that AI leadership isn’t one race. From chips and power to models and industrial deployment, global countries are positioned to capture unique value in different layers of the AI stack. Download PDF – US companies still dominate artificial intelligence (AI) capital and model production, but having an AI advantage is not a single leaderboard. In 2024, US private AI investment reac hed US$109 billion, dwarfing activity in China and the United Kingdom, which ranked as distant runners-up.1 Yet as AI investment moves from model creation to physical buildout and enterprise deployment, value creation increasingly extends beyond the United States, flowing to companies tied to global hardware supply chains, industrial systems and real-economy adoption. This shift has important implications for how investors think about global equity exposure.
Beyond TCA: How Traders Are Using Analytics to Change Behavior, Not Just Report Results – Beyond TCA: How Traders Are Using Analytics to Change Behavior, Not Just Report Results.While most buy-side firms utilize Transaction Cost Analysis (TCA), few successfully use it to drive day-to-day behavioral changes. For TCA to evolve from a “compliance artifact” into a decision-support tool, it must shift from merely reporting results to providing actionable evidence that influences broker selection, trade timing, strategy choice, and PM communication. This FactSet Insight article offers a blueprint on what’s required to take a desk’s TCA to the next level.
Anthropic suggests AI might be worse for hedge fund employees than bankers March 6, 2026. Banks are still relying on their trusty analysts to build powerpoints and financial models, but will that change as AI becomes more capable? Anthropic, developers of the AI chatbot Claude, seems to think the rumors of the death of financial services jobs are greatly exaggerated. Some of them, at least..
Peter Thiel warned AI is coming for ‘math people before word people.’ Banks have already said smaller headcounts are possible Fortune Tech, March 7, 2026.
Industry White Papers
Anthropic, March 6, 2026. Labor market impacts of AI: A new measure and early evidence. Maxim Massenkoff and Peter McCrory.
• We introduce a new measure of AI displacement risk, observed exposure, that combines theoretical LLM capability and real-world usage data, weighting automated (rather than augmentative) and work-related uses more heavily
• AI is far from reaching its theoretical capability: actual coverage remains a fraction of what’s feasible
• Occupations with higher observed exposure are projected by the BLS to grow less through 2034
• Workers in the most exposed professions are more likely to be older, female, more educated, and higher-paid
• We find no systematic increase in unemployment for highly exposed workers since late 2022, though we find suggestive evidence that hiring of younger workers has slowed in exposed occupations
PAPERS
NBER
How does AI Distribute the pie? Large Language Models and the Ultimatum Game. Douglas K.G. Araujo As Large Language Models (LLMs) are increasingly tasked with autonomous decision making, understanding their behavior in strategic settings is crucial. We investigate the choices of various LLMs in the Ultimatum Game, a setting where human behavior notably deviates from theoretical rationality. We conduct experiments varying the stake size and the nature of the opponent (Human vs. AI) across both Proposer and Responder roles. Three key results emerge. First, LLM behavior is heterogeneous but predictable when conditioning on stake size and player types. Second, while some models approximate the rational benchmark and others mimic human social preferences, a distinct “altruistic” mode emerges where LLMs propose hyper-fair distributions (greater than 50%). Third, LLM Proposers forgo a large share of total payoff, and an even larger share when the Responder is human. These findings highlight the need for careful testing before deploying AI agents in economic settings.
Chaining Tasks, Redefining Work: A Theory of AI Automation. Mert Demirer, John J. Horton, Nicole Immorlica, Brendan Lucier & Peyman Shahidi. Working Paper 34859. DOI 10.3386/w34859. Issue Date February 2026. Production is a sequence of steps that can be executed (1) manually, (2) augmented with AI, or (3) fully automated within contiguous AI-executed steps called “chains.” Firms optimally bundle steps into tasks and then jobs, trading off specialization gains against coordination costs. We characterize the optimal assignment of humans and AI to steps and the firm’s resulting job structure, showing that comparative advantage logic can fail with AI chaining. The model implies non-linear productivity gains from AI quality improvements and admits a CES representation at the macro level. Empirical evidence supports the model’s key predictions that (1) AI-executed steps co-occur in chains, (2) dispersion of AI-exposed steps lowers AI execution at the job level, and (3) adjacency to AI-executed steps increases the likelihood that a step is AI-executed.
Building Pro-Worker Artificial Intelligence. Daron Acemoglu, David Autor & Simon Johnson. Working Paper 34854. DOI 10.3386/w34854. Issue Date February 2026. This paper defines pro-worker technologies, including Artificial Intelligence, as technologies that make human skills and expertise more valuable by expanding worker capabilities. Our conceptual framework distinguishes among five categories of technological change: labor-augmenting, capital-augmenting, automating, expertise-leveling, and new task-creating. Only the last category is unambiguously pro-worker, generating demand for novel human expertise rather than commodifying it. We illustrate these distinctions through hypothetical and real-world examples spanning aviation maintenance, electrical services, custodial work, education, patent examination, and gig delivery. While AI’s capacity to automate work is substantial, we argue that its potential to serve as a collaborator, by extending human judgment, enabling new tasks, and accelerating skill acquisition, is equally transformative and currently underexploited. We identify market failures, including misaligned firm and developer incentives, path dependence, and a pervasive pro-automation ideology, that may lead to underinvestment in pro-worker AI. We consider nine policy directions that would change incentives, including targeted investments in health care and education, tax code reform, antitrust enforcement, and intellectual property protections for worker expertise.
Bank for International Settlement (BIS)
Generative AI for surveys on payment apps: AI views on privacy and technology BIS Working Papers | No 1333 | 06 March 2026 by Koji Takahashi and Joon Suk Park. The study examines whether artificial intelligence (AI) can simulate how people think about mobile payment apps, especially regarding privacy risks and perceived benefits. Using OpenAI’s ChatGPT, we generate artificial survey responses and compare them with real survey data from the Netherlands. The goal is to see if AI can reproduce realistic human attitudes, such as some people using apps despite privacy concerns while others avoid them. This highlights the broader question of whether AI can act as a substitute or support tool for traditional surveys. The paper demonstrates a new use of generative AI in survey research. It shows that AI can mimic key behavioural patterns found in real data, such as differences between users and non-users and varying privacy attitudes. It also introduces a structured method for designing AI-based surveys using personas and prompts. Importantly, it highlights how AI can help researchers test ideas, design surveys and generate preliminary insights quickly and at low cost. The results show that AI can broadly replicate human patterns: users see more benefits and fewer risks, while privacy-concerned individuals are more cautious. However, AI responses lack diversity and tend to overemphasise privacy concerns, producing biased and less varied results than real surveys. Changes in prompt design also significantly affect outcomes. Overall, AI is useful as a complementary tool, but it cannot fully replace human surveys due to limited variability and potential bias.
IMF
Working Papers
AI Meets Fiscal Policy: Mapping Government Spending Actions Across 64 Countries March 6, 2026. We build the first global quarterly narrative database of discretionary government spending actions by applying a fixed GPT–4.1 prompt to Economist Intelligence Unit (EIU) Country Reports. The resulting series identifies exogenous spending shocks—expansions and contractions—for an unbalanced panel of 64 countries from 1952:Q1 to 2023:Q4. We validate the database by (i) replicating expert narrative coding in Romer and Romer (2019), (ii) showing that the identified shocks predict subsequent movements in measured government spending, and (iii) establishing alignment with action-based consolidation measures in Adler et al. (2024). Using country-by-country VARs that treat the narrative indicator as an internal instrument, we derive the first set of comparable cumulative government spending multipliers. The median multiplier is 0.7 at horizons up to two years, with substantial heterogeneity across countries and over time. Pooled estimates imply larger multipliers in less open economies, under fixed exchange-rate regimes, and in downturns. Multipliers are smaller when uncertainty is high and larger when political support is stronger.
Federal Reserve Bank of Dallas
AI is simultaneously aiding and replacing workers, wage data suggest J. Scott Davis February 24, 2026. Federal Reserve Bank of Dallas. Artificial intelligence’s impact on the labor market will depend on whether the technology automates or augments worker tasks. Early data on employment and wages in AI-affected industries suggest it may be doing both. The distinction between codified knowledge (for example, established information gleaned from textbooks) and tacit knowledge (understanding gained through experience) is important. If AI can replicate codified knowledge but not tacit knowledge, AI will automate jobs requiring codifiable (textbook) knowledge but complement jobs demanding experiential tacit knowledge. The distinction between codifiable and tacit knowledge further suggests that AI may substitute for entry-level workers but augment the efforts of experienced workers. The data indicate that wages are rising in AI-exposed occupations that place a high value on a worker’s tacit knowledge and experience.
Employment in AI-exposed sectors lags – Total U.S. employment increased approximately 2.5 percent since ChatGPT’s release in fall 2022. However, employment trends vary significantly across sectors. Employment in the computer systems design and related services sector has declined 5 percent. More broadly, employment has declined 1 percent since late 2022 in the 10 percent of sectors most exposed to AI, according to an index developed by Edward W. Felten, Manav Raj and Robert Seamans (“Occupational, industry, and geographic exposure to artificial intelligence: A novel dataset and its potential uses,” 2021) (Chart 1)…