The Research and What It Found
A working paper from the National Bureau of Economic Research doesn’t arrive quietly. The institution has produced some of the most consequential labor economics research of the last three decades, and its methodology, a structured survey of 750 CFOs, conducted with Duke University and the Federal Reserve Banks, represents the kind of primary data collection that most AI labor market commentary lacks entirely.
The paper was released around March 24, 2026. Commentary from Jason Averbook, a workforce analyst who reviewed the paper’s findings, confirmed its existence and described its conclusions. According to reporting on the paper, it projects AI-driven job cuts in 2026 at approximately 502,000, roughly nine times the approximately 55,000 documented in 2025. These figures are from a working paper, not a final peer-reviewed publication, and they weren’t confirmed through direct access to the primary document in this reporting cycle. They warrant independent verification before being treated as established research conclusions. The NBER’s institutional credibility and the survey methodology’s rigor make the findings worth serious attention while that verification occurs.
What the headline figures don’t capture is the mechanism. Analyst commentary reviewing the paper characterizes many of the projected cuts as preemptive, workforce decisions made in anticipation of AI capabilities rather than in response to demonstrated replacement. That framing shifts the story entirely.
If companies are cutting jobs because AI already replaced the work, the displacement is reactive. It’s responsive to a proven capability. But preemptive cuts mean organizations are restructuring based on what they expect AI to do in the next 12 to 36 months. The labor market is responding to a projection, not a fact. That distinction has significant implications for how HR leaders, workforce planners, and policy designers should respond.
The Mechanism: How Expectation Becomes Action
Consider what a CFO survey actually measures. When 750 CFOs report their intention to reduce headcount in anticipation of AI-driven efficiency, they’re describing a decision already in progress, hiring freezes, non-backfilled attrition, role restructuring, and in some cases direct layoffs. The decision to act on AI expectations doesn’t require the AI to be working perfectly. It requires management confidence that the capability trajectory justifies the organizational investment.
That confidence is, itself, a market signal. The NBER paper’s survey methodology captures something employment statistics don’t: the strategic intent layer that precedes the official job cut announcement. Layoff trackers, WARN notices, and BLS employment data measure what has happened. The CFO survey measures what leadership plans to do. The gap between those two data types, typically measured in quarters, is where workforce planning actually lives.
Meta’s layoffs this week provide a real-time case study in the same pattern. NBC News reports Meta cut approximately 700 employees across five divisions, Reality Labs, Facebook, global operations, recruiting, and sales, in the same week the company announced a more than $10 billion data center commitment in El Paso, Texas. Meta hasn’t publicly attributed the layoffs to AI investment. Multiple sources frame the events together contextually. The simultaneity is the point: infrastructure spending and headcount reduction are happening in the same strategic window, whether or not the causal link is formally stated.
Who Is Most Exposed
Not all exposure is equal. The NBER paper reportedly finds a 16% employment decline among entry-level workers aged 22 to 25 in AI-exposed roles, according to reporting on its findings. That figure wasn’t confirmed through direct access to the primary document. It’s included here because, if confirmed, it’s the single most consequential sub-finding in the dataset.
Entry-level erosion is structurally different from senior-level reduction. When an organization eliminates entry-level positions, it disrupts its own workforce pipeline. Junior roles exist not only to complete work but to develop the senior talent of the next decade. Organizations that eliminate entry points, through hiring freezes, AI tool deployment, or direct role elimination, produce a workforce composition gap that becomes visible in 5 to 10 years, not in the next earnings call.
Goldman Sachs Research has estimated that approximately 300 million jobs globally are exposed to AI automation, an exposure estimate, not a displacement projection. The distinction matters. Exposure means a role has tasks that AI could perform; displacement means the human performing it is removed. Most roles with AI exposure will see task-level changes, not elimination. Entry-level roles in AI-exposed functions are where the overlap between exposure and actual displacement is highest, because those roles tend to concentrate the routine task categories that current AI systems handle most reliably.
The sector pattern isn’t uniform. The NBER survey captures CFO intent across industries, not just technology. Legal, financial services, consulting, and media organizations, all sectors with significant AI adoption in the current cycle, have meaningful concentrations of AI-exposed entry-level roles. The 502,000 projected figure is cross-sectoral; the entry-level impact is concentrated.
What This Means for HR and Workforce Leaders
Three implications are concrete enough to act on now, regardless of whether the specific figures in the NBER paper hold up under additional scrutiny.
First: workforce planning assumptions need a preemptive-displacement scenario. Most organizations model their workforce under conditions where AI adoption produces efficiency gains while headcount stays roughly stable. The NBER data, if directionally accurate, suggests the CFO cohort as a whole is not operating under that assumption. HR leaders who haven’t built a preemptive-displacement scenario into their workforce models are working with an incomplete picture.
Second: entry-level roles require explicit protection logic, not implicit assumptions. If the 16% decline figure reflects real patterns in AI-exposed sectors, organizations that eliminate entry-level positions without a deliberate pipeline strategy are creating a talent deficit that will compound over time. This requires active workforce policy, apprenticeship programs, explicit entry-level headcount floors, or structured transition pathways, rather than passive attrition management.
Third: the timing of the labor market response is earlier than most policy frameworks assume. Employment regulation, retraining programs, and social safety nets are designed to respond to displacement that has already occurred. Preemptive restructuring, by definition – happens before those mechanisms activate. Policy teams working on AI labor market response need frameworks that address the anticipatory phase, not just the confirmed displacement phase.
What to Watch
Three forward indicators are actionable for workforce and HR leaders.
Direct access to the NBER working paper is the first priority. If the full paper is publicly available through NBER’s working paper series, it should be reviewed for methodology, sector breakdowns, and whether the entry-level finding is qualified in ways the commentary doesn’t capture.
BLS employment data for Q1 2026, particularly in professional services, administrative support, and technology-adjacent sectors, will provide the first official statistical signal of whether the CFO intent translates into measured employment changes at the macro level.
Earnings calls from major technology, financial services, and consulting firms over the next 60 days will include forward guidance language that, parsed carefully, often indicates workforce intentions before they appear in official announcements. Investors and HR leaders reading these transcripts for AI-related workforce signals will have a lead indicator advantage.
TJS Synthesis
The preemptive-cuts finding reframes what the AI labor market debate has been about. The standard narrative is a debate between displacement pessimists and productivity optimists, AI will eliminate jobs versus AI will create new ones. The NBER data introduces a third dynamic: companies acting on AI expectations before the outcome is determined. That behavior doesn’t wait for the debate to resolve. It’s already in the hiring plans, the attrition policies, and the organizational restructuring decisions of 750 CFOs who collectively represent a meaningful cross-section of the economy.
HR leaders, workforce planners, and policy designers can’t respond to a displacement that’s framed as future. They need frameworks for a labor market already responding to anticipated AI, not confirmed AI. The NBER paper, pending full verification, provides the research basis for that reframing. The Meta layoffs provide the real-time illustration. Together, they suggest the AI expectation economy is operational, even if the AI economy itself isn’t complete.