At the 2026 National Two Sessions, the Beijing Municipal Human Resources and Social Security Bureau released a labor dispute case: a technology company fully shifted to AI automated data collection, dismantled its traditional data collection department, and terminated Liu's labor contract on the grounds of "objective circumstances having undergone significant changes."
The Beijing Labor and Personnel Dispute Arbitration Commission ruled: the company shifted the risks of normal technological iteration onto the worker, and the dismissal was illegal.
The truly heart-wrenching aspect of this case is not that AI took one job, but that it exposed the core flaw of the social security system: contributions have always been tied to the payroll.
When companies replace people with machines, the total wage bill shrinks, and the contribution base thins out;
when people are pushed into platforms, gig work, and self-employment, the contribution chain breaks.
Robots are equipment, artificial intelligence is a technical system; neither is a civil laborer, so of course they cannot "pay social security."
Technology and internet expert Zhang Dongwei believes that the so-called "robots paying social security" is actually asking: "Should part of the cost saved by machines replacing people be backfilled into social security?"
This is not about opening accounts for robots, but about ensuring that the bill for technological substitution cannot be paid solely by the replaced individuals and the government.
The anchor of social security has always been tied to the payroll
China's modern social security system began as a supporting project for state-owned enterprise reform. The goal was very practical: enterprises should not run society, and individuals should not be bound to their work units.
Social security is the safety net of the market economy, but the material woven into this net has always been labor relations and payroll.
The employee pension insurance combines social pooling and individual accounts, but current-period balance relies heavily on the flow of contributions from the payroll. Enterprises contribute 16% of the total wage bill, individuals contribute 8%, and the money is deducted from the payroll and goes into the fund pool.
The system's premise rests on just three conditions: enough contributors, a large enough total wage bill, and a sufficiently continuous contribution chain.
Lose one, and the pool tightens.
In 2025, the national social security fund budget revenue was 12.6 trillion yuan, of which fiscal subsidies were about 2.9 trillion yuan, accounting for about 23% of total revenue; that year, revenue still exceeded expenditure.
But this contribution structure is already worth vigilance: the growth rate of fiscal subsidies is significantly higher than that of insurance premium income.
What is more troublesome is the other end of the contribution chain. There are about 320 million flexible workers nationwide, but only about 70 million participate in employee pension insurance, a participation rate of less than 30%; of those already insured, about 38% have interrupted contributions midway.
This is not a small leak; this is the contribution base leaking.
Therefore, the erosion of the social security contribution base may not first manifest as "robots taking jobs."
More commonly: employment remains, but the contribution relationship shifts from continuous to intermittent, from employer-paid to fully individual-borne, or simply stops.
The unemployment rate does not see this part, but fund revenue does.
The old game of tying social security to the payroll has hit a wall.
Stupid! It's not the steam engine, it's computers and the internet
A friend made an extreme analogy, asking why not ask whether steam engines or cars pay social security?
To compare the impact of technological shocks on social security, one must choose an era with a social security system.
When steam engines and cars became widespread, the modern social security system had not yet been established, and there was no comparable "contribution—benefit" chain; such forced analogies are worthless.
What can truly be compared is the computer and internet era: social security already existed, technology similarly replaced jobs and created jobs, and it also left a legacy of platform employment and interrupted contributions for flexible workers.
Zhang Dongwei, who has been engaged in related work since 1995, experienced the process of computer and internet popularization and can see two lessons.
First, technological substitution did not collapse the social security contribution base as a whole.
Typists, switchboard operators, and traditional retail jobs disappeared, but IT, e-commerce, and the platform economy created new jobs; urbanization and the expansion of insurance coverage advanced simultaneously, and the contribution base overall continued to expand.
Second, it left an unresolved tail: interrupted contributions in platform employment and flexible work.
Food delivery riders, ride-hailing drivers, and freelancers are largely in a state of "having income but no stable contributions." Today's flexible workers' participation rate of less than 30% and 38% interrupted contributions are this tail.
AI and robots may amplify both lines simultaneously.
First, substitution is concentrated in standardized, codifiable jobs with a high share of the total wage bill.
Customer service scripts, ticket entry, and standard Q&A can already be handled by large models; warehouse sorting and last-mile delivery are being squeezed by mobile robots and automated systems.
And these jobs are precisely the backbone of the social security contribution base: the total wage bill is not small, participation is relatively standardized, and the contribution chain is relatively continuous.
Second, the replaced people may not be "unemployed," but more likely "low-contribution/interrupted."
The computer and internet era has already proven that total employment did not collapse, yet the contribution chain can still break at the individual level.
Current AI substitution is occurring in high-value white-collar jobs; if the substitution rate continues to advance into middle- and low-skilled white-collar and blue-collar jobs, this broken chain will spread from the platform periphery to the core of formal employment.
Therefore, in Zhang Dongwei's view, robot and AI substitution changes not only the quantity of employment but also the contribution quality of employment.
The contraction of the contribution base occurs within the structure of participation, not simply in total employment.
The computer and internet era generally preserved the contribution base but did not solve the interruption problem; what AI must answer is whether it will turn this legacy problem into the main problem.
Enterprises calculate small accounts: machines replacing people saves money;
society must calculate the big account: part of the saved money must return to social security.
How to fix the system
When "robots paying social security" lands in policy, the first thing that comes to mind is a "robot tax."
This concept is not new and has no consensus. In 2017, the European Parliament discussed giving robots legal personality and taxing owners, but the proposal was rejected. In 2018, South Korea reduced the tax credit for automation investment by large enterprises from 7% to 3%, which outsiders called a "quasi-robot tax"; some studies say that afterward, South Korea's industrial robot installations declined relative to Japan.
Currently, no country has formally legislated a narrow "robot tax."
This topic is not something that cannot be discussed; rather, don't keep fixating on physical robots.
Taxing robots is hard to define: what counts as a robot, and does software automation count?
Moreover, this would also create double taxation: enterprises have already paid VAT and income tax when purchasing and using equipment.
1. What may be more operable is to focus on the "social costs saved by machines replacing people" and build compensation channels.
At the 2026 Two Sessions, representatives have already proposed establishing a "Special Fund for Technological Transformation" to provide job-transfer subsidies, resettlement allowances, and reemployment rewards for employees in replaced industries and enterprises, and to appropriately adjust the tax burden on enterprises that are highly automated and replace large numbers of people.
This is closer to China's policy context than a narrow-tax-base "robot tax."
2. Stop the contribution chain for flexible employment from breaking again.
The key is not to shout "expand coverage," but to lower the threshold and allow flexible contributions.
When flexible workers participate in employee pension insurance, they usually bear the full 20% of the average social wage personally, while only 8% goes into the individual account; in large and medium-sized cities, the annual contribution burden generally exceeds 10,000 yuan.
If the burden is not genuinely reduced, expanding coverage will become "expanding coverage but making it hard to avoid interrupted contributions."
The legacy of gaps left by the computer and internet era must be prioritized for repair in this round of technological replacement; it cannot be delayed any longer.
3. Stock supplementation and flow reconstruction should proceed in parallel.
The transfer of state-owned capital to replenish social security funds has been running for years, addressing historical debts and stock gaps.
Feeding automation gains back into the system must address the flow problem: as value creation shifts from labor to capital, data, and technology, the social security contribution base must also follow the structural shift.
Bringing part of the incremental gains from automation into social security through a broad tax base is more reliable than creating a new narrow tax category targeting "robots."
Dongcha
Robots do not pay social security. But the costs saved by machines replacing people cannot all become pure corporate profit;
the money for the social safety net must be recovered anew from the new mode of production.
First, robots are not contribution entities, but the social costs of automation replacement cannot all be borne by replaced workers and fiscal subsidies.
In 2025, fiscal subsidies already accounted for about 23% of the budgeted revenue of social security funds; subsidies can buy time, but they cannot replace contribution base construction.
Second, social security financing should shift from "payroll dependence" to a "diversified contribution base."
The declining support ratio within the system, less than 30% participation among flexible workers, and 38% interrupted contributions all point to one thing: the stability of the payroll as the sole anchor is declining.
The computer and internet era preserved the contribution base, yet left gaps in platform employment; if AI advances into the core of formal employment, this legacy will become a mainline problem.
Third, the most operable starting point in the short term is not a "robot tax," but three parallel measures: build a compensation mechanism for the social costs of machines replacing people; substantively lower the contribution threshold for flexible employment, making "non-interruption" a priority over "coverage expansion"; and use a broad tax base to bring part of the incremental gains from automation into social security financing.
Let the contribution base follow value creation; do not cling to the payroll.
Let the gains and costs of technological replacement be realigned onto the same social contract.
This article is from the WeChat public account "Zhang Dongwei," author: Zhang Dongwei
