AI Is Redrawing the Class Map, Not Just in China: Anthropic Data Show White-Collar Roles Face the Sharpest Shock

A Chinese debate meets a global labor reset

Huxiu (虎嗅), a prominent Chinese tech and business outlet, argues that China’s decade-long online fixation on “class solidification” misses a more urgent rupture: an AI-driven split that is already reshaping who wins, who loses, and how fast. The premise is simple yet disruptive—your position in the social hierarchy is no longer anchored in wealth, pedigree, or diplomas, but in your proximity to AI tools. That message lands in a country where “upward mobility” and “inherited advantage” have fueled relentless debate, yet the underlying labor market shock is global, not uniquely Chinese.

What the new data show

On March 5, Anthropic published “Labor market impacts of AI,” introducing an on-the-ground metric—Observed Exposure—to track what AI is replacing in real workflows. The headline numbers invert old assumptions: software developers top the list at a 74.5% observed exposure rate, followed by customer service representatives (70.1%), data entry clerks (67.1%), medical records specialists (66.7%), market research and marketing analysts (64.8%), and financial and investment analysts (57.2%). An Anthropic chief engineer has reportedly said that almost all of the company’s code is now written by AI. This aligns with OpenAI’s 2023 “GPTs are GPTs” paper, which estimated that 19% of U.S. workers could see over half their tasks affected within a decade—an adoption curve that Cognizant now says arrived roughly six years early, with roles facing 50%+ task exposure jumping from near zero in 2023 to about 30%, and average annual exposure growth accelerating from 2% to 9%.

Beyond coders: management, blue-collar spillovers, and who is most exposed

The blast radius is widening. Cognizant’s update estimates CEO-level AI exposure has climbed from around 25% to more than 60%. Multimodal models and AR wearables are lifting exposure in construction (from roughly 4% to 12%) and transportation (about 6% to 25%). Even so, the gap between what AI could do in theory and what it is doing in practice remains large: for “computer and math” jobs, theoretical coverage approaches 94%, yet observed use sits near 33%, suggesting a substantial, not-safe-yet-to-be-contested frontier. Demographically, Anthropic’s analysis indicates that the quarter of workers with the highest exposure skews higher-paid (average $32.69/hour, about 47% above the low-exposure cohort), more likely to hold graduate degrees (17.4% vs. 4.5%), with a higher share of women (+16 percentage points), more white workers (+11 points), and a near-doubling share of Asian workers—undercutting the notion that only low-skill roles are at risk.

Why this matters for China—and beyond geopolitics

For Western readers, China’s “class solidification” debate centers on fears that family resources trump effort. Huxiu’s argument reframes the issue: AI is compressing “average” cognitive output into a commodity, eroding middle-class moats built on licenses, processes, and experience. That shift collides with China’s own AI race, as Baidu (百度), Alibaba (阿里巴巴), Tencent (腾讯), and ByteDance (字节跳动) push large models across search, cloud, and consumer apps—moves unfolding under U.S. export controls on advanced chips that pressure domestic AI stacks. The stakes are clear: high-skill white-collar roles in software, finance, marketing, and customer service—the backbone of China’s urban middle class and a mirror of advanced economies—are being redefined at speed. Compete with the machine on tasks—or learn to design, deploy, and direct it? That, Huxiu suggests, is the real sorting mechanism in the AI era.