While public debate over artificial intelligence fixates on mass unemployment and dystopian job losses, regulators and economists are quietly documenting a different problem: companies exaggerating what their AI actually does. The U.S. Securities and Exchange Commission has already penalized firms for misleading investors about AI capabilities, and new economic research shows that real-world AI adoption is producing targeted labor shifts rather than the sweeping displacement that dominates headlines. The gap between the panic and the evidence matters because it shapes policy, investment decisions, and how workers prepare for what comes next.
Regulators Are Chasing AI Hype, Not AI Harm
The SEC charged two investment advisory firms, Delphia (USA) Inc. and Global Predictions Inc., with making false and misleading statements about their use of artificial intelligence. Delphia claimed it used AI and machine learning to analyze client data and predict market performance in ways the firm could not actually deliver. Global Predictions marketed itself as the “first regulated AI financial advisor” and overstated how its technology worked. Both firms settled the charges, paying $400,000 in combined civil penalties, a modest sum in financial-market terms, but a clear warning shot for the rest of the industry.
The cases illustrate a pattern that regulators have labeled “AI washing,” a term borrowed from greenwashing in the environmental space. Companies attach the AI label to products and services to attract capital and media attention, even when the underlying technology falls short of the marketing. For investors and consumers, the risk is not that AI will suddenly take their jobs but that they will make financial decisions based on capabilities that do not exist. The SEC enforcement actions signal that the regulatory apparatus is, at this stage, more concerned with policing false promises than with constraining the technology itself. That suggests that the most immediate harms stem from misrepresentation rather than runaway automation.
Firm-Level Data Shows Substitution, Not Collapse
A working paper posted on arXiv offers one of the first firm-level looks at how companies actually redirected spending after generative AI tools became widely available. The researchers used payments data spanning Q3 2021 through Q3 2025, treating the late-2022 rollout of widely accessible chatbots as an adoption shock. Using a difference-in-differences design, they found that firms reduced spending on contracted online labor marketplaces while increasing payments to AI model providers. The pattern is real, measurable, and specific to certain categories of outsourced work, particularly digital tasks that translate easily into promptable workflows.
Yet the shift looks nothing like the wholesale replacement of human workers that dominates public fear. The substitution effect the researchers identified was concentrated among tasks already being outsourced to freelance platforms, not across entire payrolls or industries. In practice, companies appear to be swapping one vendor cost for another in targeted areas rather than eliminating headcount at scale. That distinction matters enormously for workers trying to assess their own exposure: freelancers who rely on repetitive writing, basic coding, or simple design gigs face more immediate pressure than in-house staff. For policymakers, the findings argue for narrow interventions (such as support for platform workers and skills upgrading), rather than sweeping assumptions that all jobs are equally at risk.
Productivity Numbers Tell a Quieter Story
If AI were truly transforming the economy at the speed the panic suggests, the productivity data should show it. The U.S. government’s official time series for labor productivity, output, hours worked, unit labor costs, and compensation in the nonfarm business sector, published regularly by the Bureau of Labor Statistics, show steady but unremarkable gains, not the kind of explosive acceleration that would indicate a technology-driven restructuring of how goods and services are produced. Recent readings for 2024 and through the third quarter of 2025 fall within historical ranges, with quarter-to-quarter wiggles that look more like normal business-cycle variation than the footprint of a revolutionary tool suddenly boosting output per worker.
Short-term productivity figures also carry significant noise, which complicates attempts to draw sweeping conclusions about AI from a single data point. The BLS maintains an archive of preliminary and revised quarterly releases, and its historical revisions show that initial estimates often change meaningfully as more information comes in. Anyone claiming that one quarter proves AI is supercharging or destroying the economy is likely cherry-picking from data that will look different after revision. The more cautious reading is that AI integration is happening, but its macroeconomic footprint remains modest enough that it blends into the usual statistical noise. For citizens, journalists, and researchers, the public data tools provided by the BLS make it possible to track these trends directly instead of relying on breathless commentary that may overstate both the promise and the peril.
Geography Flips the Automation Script
Previous waves of automation hit hardest in rural, manufacturing-heavy regions where robots and software replaced routine factory and back-office tasks. Generative AI is doing something different. According to the OECD’s 2024 analysis of job creation and local economic development, task-level exposure to generative tools is often higher in metropolitan, service-heavy regions and among high-skilled workers. Urban professionals in finance, law, marketing, and software development face greater exposure to partial automation than factory workers or agricultural laborers, because their daily tasks involve information processing, drafting, summarizing, and analysis that current models can assist with.
This geographic inversion has direct consequences for how governments should respond. Retraining programs designed for displaced manufacturing workers in small towns do not translate neatly to white-collar professionals in major cities whose tasks are being partially automated rather than eliminated. The OECD’s comparative figures for urban versus non-urban areas suggest that generative AI reverses patterns seen in earlier automation waves, which means the policy toolkit needs to be rebuilt rather than reused. City governments and national agencies will need to think more about mid-career reskilling for lawyers, analysts, and designers, and less about the old assumption that “knowledge economies” are automatically insulated. At the same time, regions that bore the brunt of previous technological shifts may find themselves relatively shielded in this round, complicating long-standing narratives about who is most vulnerable to automation.
The Real Risk Is Misdirected Attention
The evidence from regulators, firm-level research, government statistics, and international policy analysis converges on a single point: the loudest version of the AI story is also the least accurate. Jobs are not vanishing overnight. Productivity is not surging beyond historical norms. The measurable changes are real but targeted, affecting specific types of outsourced work and particular categories of high-skill urban employment. The actual threat that deserves more scrutiny is the growing gap between what companies claim their AI can do and what it actually delivers, a gap that the SEC has already started to confront through its early enforcement actions on misleading disclosures.
Refocusing attention means using the tools and institutions that already exist. The U.S. Department of Labor oversees labor standards, training initiatives, and worker protections that can be adapted to address AI-related shifts without assuming an imminent jobs apocalypse. Economists can continue to mine transaction-level data to identify where substitution is occurring and which workers are most exposed, rather than extrapolating from speculative scenarios. Local and national policymakers can draw on international comparisons to understand how geography and skill level shape exposure, instead of defaulting to outdated narratives from past automation waves. The real risk is not that AI will instantly upend the labor market, but that public debate will remain fixated on extreme scenarios while overlooking the slower, more targeted changes that are already underway, and the very real problem of AI being sold as something it is not.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
