How some Chinese data labeling companies hired by Baidu, Alibaba, and JD.com to train AI are exploiting vocational school students via data labeling internships
the exploitation of cheap labour in the AI industry; often underpaid and overlooked critical roles 1/2 https://restofworld.org/... Viola Zhou / @violazhouyi : Tech companies like Baidu have established data annotation centers inside vocational schools. But researchers say such internships expose students, many of whom from lower-class or rural backgrounds, to labor exploitation and abuse. READ MORE https://restofworld.org/... Yi-Ling Liu / @yilingliu95 : China's AI boom depends on an army of exploited student interns https://restofworld.org/... As part of China's digital underclass, vocational school students work as data annotators - for low pay & poor prospects. @violazhouyi & @CaiweiC 's latest investigation for @restofworld Vicki Turk / @vickiturk : I've seen plenty stories about AI labor exploitation but this is new - students being made to annotate data in order to graduate https://restofworld.org/...
Context & Ripple Effects
This report places vocational schools inside a longer Chinese data-work supply chain: earlier coverage described low-cost labeling of images and surveillance footage in China, while this account identifies student internships as a channel for that work.
It also fits a broader AI-production pattern in which annotation demand is met through workers with limited bargaining power, as seen in data-labeling firms sourcing labor in crisis-hit economies. The significance is that the labor layer behind model development can be organized through educational institutions as well as outsourced platforms.
First-order effects
- Vocational students reportedly perform data-labeling work through internships under conditions that can mean low pay, weak choice, and exposure to abuse; the immediate cost advantage accrues to the labeling suppliers using them.
- Baidu, Alibaba, and JD.com are tied to a data supply chain in which annotation centers inside schools can make large pools of trainee labor available to their contracted vendors.
Second-order effects
- Enterprise buyers and labeling vendors may face sharper scrutiny of internship terms, school partnerships, and responsibility for working conditions deeper in their training-data supply chains.
- If schools become a contested source of annotation labor, vendors could need to compete more for adult workers or make labor practices more visible, reducing the appeal of opaque, low-cost sourcing.
Third-order effects
- The case points to AI infrastructure becoming institutionalized beyond cloud and chips: education systems can become part of the recurring labor machinery needed to produce training data.
- If similar reporting persists across markets, procurement and labor accountability may increasingly extend from model developers to the intermediaries and institutions that supply human data work.
The trend: AI’s expansion is turning data labeling into a strategically important but weakly protected labor layer, with institutions such as schools becoming potential channels for scaling it.