Imagine losing your job to a system your employer spent the previous year asking you to help improve. You apply elsewhere. The other companies have bought something similar. There is work, but less of it, and a queue of people willing to do it for less.
Then the government announces a monthly payment. Enough for groceries, electricity, most of the rent. You register, the money arrives, and for the first time in months you sleep properly.
Of course you take it. You have children. Whatever reservations you have about the people offering it can wait until the fridge is full.
That is where I would begin thinking about universal basic income: with the relief. Any serious account of how it could become a means of control has to understand why people would welcome it, and why accepting it would be an entirely reasonable thing to do.
Table of Contents
Who benefits when wages disappear
By late 2026, the question has moved beyond whether AI can draft a document or generate code. It is whether companies will keep hiring people to do the work. A Stanford payroll study updated in August 2026, using data through June, found employment among 22–25-year-olds in highly AI-exposed occupations about 19% below where it would have been had it kept pace with similarly aged workers in less-exposed occupations. The adjustment appeared to come mainly through reduced hiring. The door was closing before people had a chance to enter.
The researchers found no widespread, economy-wide displacement, and the comparison does not establish how much of the gap AI caused. It would be dishonest to turn it into a claim that a fifth of all jobs has disappeared. But it describes a danger that a headline unemployment rate can obscure: an economy can continue employing its established workers while offering fewer places to the people coming after them.
But the danger does not require the last plumber to be replaced by a robot. It requires enough people to lose reliable income that refusing the government's alternative becomes impractical. A society can reach that point with plenty of jobs still advertised.
The companies building AI are already helping frame the response. OpenAI's April 2026 policy paper proposed a Public Wealth Fund that could distribute investment returns to citizens. In his new policy essay, Dario Amodei argues that permanently reduced demand for labour could require long-term income support, including UBI financed by taxes on companies or capital gains. These remain proposals, with different implications for ownership. Their significance is that the firms selling the disruption are also becoming influential participants in the debate over how everyone else will live with it.
Still, there is a commercial interest here that deserves attention. Companies need customers. An economy that eliminates wages faster than it replaces them eventually has a sales problem. Public payments can keep purchases flowing while allowing the companies doing the automating to remain profitable. They can also make the disruption politically tolerable. A government cheque is easier to sell than a prolonged argument over who should own the machines.
That leaves a possibility worth following through: the public receives enough income to keep consuming, while control over production, housing, infrastructure and prices remains concentrated elsewhere.
Under those conditions, the amount on the cheque tells you surprisingly little about how free its recipient is.
When survival depends on one payment
Today, a basic income could give someone the means to leave an abusive employer, turn down degrading work or start a business. That is a substantial argument in its favour. But consider the version of the future in which human labour is no longer needed at anything like its present scale. In that economy, leaving one employer for another becomes a much weaker escape route. So does threatening to withhold your labour. The business can carry on without you.
If the monthly transfer becomes your only dependable income, whoever controls it acquires a power over your life that a competitive labour market, however imperfectly, spreads among many employers. The law may call the payment an entitlement. You still have to live through what happens when the office administering it changes the rules.
Start with a budget squeeze. The government promises to preserve the payment, but says the public needs reassurance that it is being used responsibly. Part of it becomes a housing credit. Another part is reserved for food and utilities. Approved providers offer discounts, so signing up with them makes the allowance stretch further.
Each change has a defensible explanation. Together, they reduce your choices. You can receive the full nominal amount and still be unable to move somewhere cheaper, buy from an unapproved seller or save enough to leave.
Purpose-restricted digital payments already exist. India's e-RUPI is a voucher delivered by SMS or QR code, designed for a particular recipient and purpose. It can make services easier to access. It also demonstrates that restricting where a benefit goes requires no futuristic currency. Whether such restrictions belong on a person's entire living allowance is a political decision.
Landlords and essential-service providers have incentives of their own. Where competition is weak and supply is constrained, some of a larger allowance can be absorbed by higher charges. If the response is another subsidy tied to approved providers, recipients become more dependent on the same businesses. The state pays; the businesses collect; the household has little room to negotiate with either.
The conditions arrive quietly
Now add the politics of deservingness.
People who welcomed emergency support can become resentful of someone they believe is receiving it too easily. A story about fraud travels further than a patient explanation of why universality reduces bureaucracy. Politicians know this. So do newspapers looking for an outrage and contractors selling fraud detection.
The first conditions could be modest: keep your details current, complete a training course, attend a review. Later, a government might add compulsory placements or approved civic activity. Even in an economy supposedly running out of work, officials can find ways to demand proof of effort. Anyone who objects can be accused of endangering the programme for everyone else.
Once behavioural conditions arrive, the payment has ceased to be unconditional. Its name may take longer to catch up. By then, families may have arranged their entire lives around receiving it.
Surveillance becomes cheap enough
Enforcing those conditions used to involve a great deal of human attention. That constraint is weakening.
Cameras, transaction records, location data and government files already contain more information than officials can personally examine. AI lowers the cost of making it searchable and acting on it: transcribing speech, translating posts, matching images, summarising records, flagging patterns. In the same policy essay, Amodei warns that AI will make bulk analysis of personal information bought from private companies much more revealing and useful for surveillance. The capacity to collect the data was already there. The human effort needed to interpret it is becoming less of a constraint.
The next step is software that acts on what it finds. Stanford's 2026 AI Index, reviewing recent advances, reports roughly 66% task success for leading computer-using agents on OSWorld, a benchmark of tasks across operating systems. That still means failure on about one attempt in three. It nevertheless points towards administrations in which software can read a file, operate an application and prepare an action for approval. For a benefits recipient, the question becomes how much scrutiny that approval actually receives.
A government no longer needs a clerk to read every post before it can attempt to monitor a population. It can run automated screening and send selected cases for investigation. Accuracy may be poor. For the person incorrectly flagged, that offers little comfort if the flag holds up next month's payment.
The connection between surveillance and punishment is already visible. In a May 2026 investigation of Iran, Amnesty International documented preferential internet access tied to registered SIM cards and identity verification, alongside messages threatening people identified through their digital activity with the loss of phone services. This does not establish that AI made the identifications. It establishes the governing principle: access to an everyday necessity can be made dependent on a political classification. More capable AI would give authorities further means to enforce it.
Where the state and its contractors meet
Under a dictatorship, the purpose can be explicit: identify disobedience and make it costly. In a democracy drifting towards authoritarianism, the same capacity can grow through fraud prevention, public-order measures and data-sharing agreements. A later government inherits whatever the earlier one installed.
Britain offers a narrower, concrete example of how the connections are being built. DWP guidance published on 13 October 2026 explains powers requiring banks to check accounts receiving certain benefits, including linked accounts, against eligibility indicators and report matches. The stated limits matter: no real-time account access, no spending histories or political opinions supplied under these checks, and entitlement decisions made by staff. This is not political scoring. It is a current example of welfare administration extending into bank screening under an anti-fraud justification.
A benefit registry gives those records a point of attachment: the identity of the person being paid. Linking it to policing or commercial risk data would require decisions, contracts and, in some places, changes to law. Those are the decisions to watch, especially when connecting another database is presented as a routine measure to protect public money.
There is an industry ready to help. In March 2026, Microsoft urged benefits agencies to treat eligibility as a “continuous, connected process”, supported by identity and shared data across programmes. Its pitch promises better services and human accountability; it is not evidence of automated punishment. It does show how continuous assessment becomes a product governments can buy.
A vendor wants a larger contract. A department wants fewer improper payments. A minister wants a system that appears decisive. Each can pursue its own interest while producing a system that gives the state more power over the individual. Their ambitions need not be identical for the machinery to serve them all.
Responsibility becomes conveniently scattered. The software company supplies a risk assessment. The agency sets the policy. The payment provider applies the hold. When you ask who decided that you should miss the rent, each can point somewhere else.
Social credit without a public score
This is how a basic-income system could acquire the function of social credit without ever introducing a public score.
One database records a missed appointment. Another flags a financial association. A public-order record enters an eligibility review. An insurer, landlord or benefits office receives a risk classification without the underlying explanation. The systems need only pass consequences between them. A single number beside every citizen's name would be optional.
Imagine, now, that the person receiving the allowance wants to attend a demonstration. There is no published rule against it. But someone they know had their account placed under review after being arrested at a protest. The money eventually came through. The landlord did not wait.
You do not have to believe every rumour to make the calculation. You only have to be unable to afford the experiment.
That is where control begins to operate without an order. You leave the group chat. You stop donating. You decide that this particular protest can happen without you. At home, you explain that you are thinking of the children, which is true.
A system that makes people cautious about exercising their rights has achieved something significant even if it rarely imposes the maximum penalty. Uncertainty helps it. You cannot confidently avoid a boundary you are not allowed to see.
Nor does the punishment have to be permanent. A two-week delay, an extra verification step or an appeal that takes longer than your remaining savings can discipline a household. Wealthy people can wait for a mistake to be corrected. People living from one transfer to the next have to accommodate the institution making it.
This would be a form of political captivity. People could remain legally free to speak, organise and refuse, while the practical cost of doing so was attached to food and shelter. Describing it only as a privacy problem would miss the extent of the power involved.
Could you still afford to say no?
There is no need to assume that every supporter of UBI wants this outcome. The more troubling possibility is that a programme introduced by people who sincerely mean to help becomes indispensable, then passes into less scrupulous hands. The beneficiaries cannot suspend their need to eat while they contest what has changed.
That is why the argument cannot end at the promise of a monthly amount. It has to reach ownership, housing, competition, the right to use money without behavioural restrictions, and the power to challenge a suspension before it causes harm. A payment backed by enforceable rights and independent institutions is different from an allowance administered at the discretion of a government and its contractors.
AI could make it possible to provide a decent life with far less compulsory work. But if the machines, the necessities and the systems distributing access remain controlled by a small number of institutions, people may be freed from employment only to become dependent on permission.
The question I would ask of any promised basic income is whether I could accept it, openly oppose the government providing it, and still be certain of eating next month.
If the answer depends on staying in good standing, the price is obedience.