AINEWS
A Stanford economist says the 20% of workers most exposed to AI saw a smaller unemployment rise since 2022 than the least exposed, and compares the pace to the decades long computer revolution.
Since the launch of ChatGPT in late 2022, unemployment for the 20% of workers most exposed to AI has risen 0.77 percentage points. A Stanford Institute for Economic Policy Research (SIEPR) brief finds that is less than the 0.85-point rise for the workers least exposed to the technology. The comparison cuts against a year of CEO warnings that the AI buildout would erase the entry-level white-collar workforce.
A year ago this month, Anthropic CEO Dario Amodei said "half" of all entry-level white-collar jobs would vanish within five years. OpenAI CEO Sam Altman, a month later, predicted the end of "certain job categories." Companies began citing AI in their layoff memos. Workers organized. Students reconsidered their majors. Then the labor data arrived, and the carnage didn't.
The SIEPR brief, summarized this week by The Guardian, is the most rigorous effort yet to test those predictions. Researchers ranked occupations by their exposure to AI and tracked unemployment in the most- and least-exposed groups. SIEPR fellow Erika McEntarfer, a co-author, said employment trends in the occupations expected to show AI impact first are "largely stable" and compared the pace to the decades-long computer revolution, in which fears of job loss preceded measurable changes by years.
The 0.77-vs-0.85 gap is small in absolute terms: both groups are still close to a 4% unemployment rate, and the SIEPR team calls it a snapshot, not a verdict. It is, however, enough to invert the public narrative. The workers whose jobs were supposed to vanish first aren't the ones losing them fastest. If anything, they have weathered the past three years marginally better than workers in occupations with little AI exposure.
The single labor-market signal that has alarmed observers is in a different place: recent graduates. Earlier in 2026, the unemployment rate for new college graduates reached 5.6%, against a national average of 4.2%. The SIEPR analysis attributes part of the gap to the unwinding of pandemic-era overhiring in tech and finance, and to the rapid expansion of remote work, which flattened the geographic advantage that had pulled graduates into big-city jobs. AI is one factor in the graduate figure, not the dominant one.
The measurement problem is real. Government statistics track occupations, not technologies, and they lag the labor market by months. Private payroll data is more current but less comprehensive. A handful of private firms now sell AI-exposure scores, but they are proprietary, and the field hasn't settled on a single definition of "exposed." So the question of whether AI is in fact eating entry-level work is being asked of instruments that were built for a different question.
The CEOs who issued the warnings have begun to soften them. Amodei and Altman now say AI will augment workers rather than replace them, and the firms they run are restructuring around that idea. Hiring managers say they increasingly expect job candidates to demonstrate AI skills, and the shape of the entry-level job is changing in subtler ways than the warnings predicted. More of the new work, recruiters and labor economists say, is moving into freelance and contract arrangements as firms try to figure out which skills they need and for how long.
For workers and students, the more useful frame is the one McEntarfer reached for: the computer revolution, not the industrial one. The personal computer arrived in offices in the early 1980s. The occupations it most threatened (clerks, typists, bookkeepers) didn't collapse for another two decades, and in some cases didn't collapse at all. The displacement happened, but on a longer, messier timeline, and it was spread across cohorts rather than concentrated in a single generation.
The Stanford finding isn't a clean exoneration of the AI-displacement thesis. The exposure rankings are a rough instrument, the time window is short, and most of the AI tools reshaping white-collar work arrived after the data the brief relies on. A year of labor data hasn't vindicated the warnings. A longer, slower timeline is the more honest base case for anyone planning a career around what AI can and cannot yet do.