AI is speeding up the deadline for basic income

UBI was already justified. AI is now creating its own independent reason to act faster.

Basic income was always justified because it is a foundation for human flourishing. Whether the economy is booming or falling does not change that principle. The rise of AI is compounding the justification because it will make an economy that provides vast wealth to some while alienating the majority.

Basic income advocates have always had to be careful with arguments related to automation. During the 2020 Democratic primary debates, I remember thinking candidate Andrew Yang was too early using automation as a justification for UBI. More importantly, framing basic income in this way distracts from the philosophical justification.

The strongest case for basic income never needed a prediction that paid work would vanish, and that caution was right.

But caution is not denial. At some point, refusing to update becomes its own kind of hype in the opposite direction. The serious question is no longer whether AI will instantly eliminate all work — it will not. It is whether AI is reorganizing the labor market faster than existing welfare states can handle. The answer is increasingly yes.

On June 4, Challenger, Gray & Christmas reported that U.S. employers announced 97,006 job cuts in May 2026, the highest May total since 2020. Employers cited artificial intelligence for 38,579 of them — 40 percent of all cuts that month, up from 7 percent in January. It was the highest monthly figure ever recorded for AI as a layoff reason since the firm began tracking it in 2023, and the third straight month AI led every other cause. For the year so far, AI has been cited in 87,714 cuts, already well past the 54,836 attributed to it in all of 2025.

That does not mean AI caused every one of those losses; companies have every incentive to use “AI” as a clean public explanation for ordinary cost-cutting. But the excuse is revealing. A firm does not blame layoffs on AI unless executives, investors, and workers already read AI as a restructuring force. Tool or pretext, management is now reorganizing work around it.

The same day, Anthropic published “When AI builds itself,” a warning about recursive self-improvement. The company said it is delegating a growing share of AI development to AI systems themselves; that its engineers now ship roughly eight times as much code per quarter as they did from 2021 to 2025; and that more than 80 percent of the code merged into its codebase is now attributable to Claude. It stressed that recursive self-improvement is not here and not inevitable but warned it could arrive sooner than most institutions are prepared for.

Set the two side by side. At one end of the economy, managers are using AI to justify record layoffs; at the frontier, engineers are using AI to speed up the building of AI itself. Neither report causes the other, but together they widen the range of futures a welfare state has to plan for — and that changes the policy calendar.

The next morning, June 5, the Bureau of Labor Statistics reported that the broad labor market was holding: total nonfarm payrolls rose by 172,000 in May, and unemployment was unchanged at 4.3 percent. Anyone claiming AI has already produced mass unemployment is overstating the evidence. Even Andy Challenger, whose own firm tallied the record AI layoffs, called it “not yet the jobpocalypse some predicted.”

But waiting for economy-wide unemployment to spike before acting would be a policy failure. AI disruption may not arrive first as breadline unemployment. It may arrive as fewer entry-level jobs, slower hiring into exposed occupations, reduced hours, compressed wages, contractor instability, and a quiet shift in bargaining power from workers to firms.

Anthropic’s own labor-market research is cautious on exactly this point. It found no clear unemployment effect yet for the most AI-exposed occupations, but tentative evidence that workers aged 22 to 25 are entering those professions at lower rates. That is the kind of early warning policymakers usually miss (or ignore), because it looks less like a crash and more like a missing rung on the career ladder.

The old welfare state was built to respond after a worker loses a job. AI may weaken people before that moment: fewer interviews, lower freelance rates, fewer junior openings, more unpaid retraining, more pressure to accept whatever terms remain. If the harm is diffuse, delayed, and hidden inside “productivity,” unemployment insurance alone will not catch it.

The public conversation keeps getting distracted by the wrong question: will AI become humanlike? That is not the threshold that matters for workers. AI does not need consciousness, emotions, or science-fiction personhood to change the economy. It only needs to perform enough economically valuable tasks cheaply enough that companies reorganize around it.

That is already happening. Anthropic’s Economic Index found that about 49 percent of jobs have seen at least a quarter of their tasks performed using Claude, up from 36 percent in early 2025 — and the work is migrating from casual chat interfaces toward API and agentic workflows that look more like automation than ordinary assistance.

The first AI labor shock is rarely a robot replacing a whole worker. It is a manager asking why a team of ten cannot become a team of six, a company deciding not to hire juniors because senior staff with AI tools can stretch further, and the slow collapse of the pipeline that used to turn beginners into experts. The worker stays employed but now competes against a machine that never asks for health insurance, rent, sleep, or dignity.

Even if recursive self-improvement is not imminent, the direction is clear: AI is increasingly used to build, test, debug, and accelerate AI. That creates a compounding force. If AI helps build better AI, the pace of change becomes less tied to human institutional speed and more tied to compute, capital, data, energy, and competition.

Anthropic itself sketched a future in which humans play a diminished role in AI development, shifting toward oversight and verification of a growing virtual lab run by AI systems — and admitted it is hard to predict what the economy looks like if human labor stops being competitive. A leading AI company is effectively saying it is not sure whether the economy can absorb what it is building, or whether institutions are moving fast enough. Then build the floor now.

The strongest basic income argument remains human, not technological. People need security because life is unstable, and bargaining power because employers, landlords, creditors, and bureaucracies often hold too much control over daily survival. They need room to leave abusive relationships, care for children, recover from illness, study, relocate, or simply breathe. That was true before ChatGPT and would remain true if the boom stalled tomorrow.

What AI changes is the risk profile. Conditional welfare asks the state to decide who is poor enough, unemployed enough, disabled enough, or deserving enough — and AI disruption will not respect those categories. A designer may lose contracts without ever appearing unemployed; a junior lawyer may never be hired in the first place; a teacher may keep the job but be expected to produce twice as much for the same pay. A universal cash floor is not perfect, but it is built for uncertainty: it reaches people before the form of harm is legible to the bureaucracy.

Reskilling matters, but it is not enough. The old answer to technological change was education: learn the new tool, move up the value chain, become more adaptable. That is useful advice at the individual level, but it is not a social contract.

The problem is that AI is climbing the skill ladder with us. When automation threatened factory work, the advice was to move into knowledge work; when AI threatens knowledge work, the advice becomes to supervise AI. But not everyone can become an AI manager, and even AI managers can be made productive enough that fewer are needed. A society cannot build economic security on an endless game of musical chairs in which every worker is told to outrun the next model release. Basic income does not replace education — it makes education realistic, giving people time to retrain without panic, reject bad jobs, and take risks without falling through the floor.

If reskilling alone cannot carry the weight, what can? The basic income world is already testing one answer. In March 2026, BIEN reported on an “AI Dividend” run by two nonprofits, the AI Commons Project and What We Will, sending $1,000 a month for a year to a cohort of 25 to 50 workers who had lost pay, jobs, or opportunities to AI, on about $300,000 in initial funding with hopes of distributing $3 million in 2026.

The revealing part is how its organizers plan to scale: by lobbying profitable AI companies to donate. That is the right instinct — the wealth AI creates should reach the people it disrupts — wired the wrong way, as charity that depends on the goodwill of the firms doing the disrupting. This is where the debate should go next. Not sympathy for a handful of displaced workers, and not a temporary patch for those who can prove AI harmed them, but a standing public claim on AI-era wealth, paid out in cash. The right language is joint ownership instead of simply compensation.

AI is not built out of nothing. It rests on public research, public universities, open scientific knowledge, government procurement, legal systems, energy grids, water systems, semiconductor supply chains, publicly educated workers, and a vast archive of human culture. Then the gains concentrate: model companies, cloud and chip firms, data-center owners, and major shareholders capture the first wave of wealth, while workers are told that disruption is the price of progress and families are told to be flexible.

If AI creates a productivity windfall, the public should receive part of it automatically. If AI creates labor disruption, the public should receive protection automatically. And if AI companies depend on public infrastructure and public knowledge, the public should hold a permanent claim on the upside. The cleanest version of that claim is an AI Social Dividend Fund. Bernie Sanders’ proposed 50 percent one-time tax on AI stocks is ethically justified, this is likely to be too economically distorting to maximize the upside for the public.  

A practical AI Social Dividend Fund

The design should be simple enough to explain and hard enough to corrupt.

Why universal beats targeted in the AI era

Targeting sounds efficient until the world changes faster than the eligibility rules. Who counts as “AI-displaced”? The worker laid off after a company adopts AI? The freelancer whose rates collapse? The graduate who never gets the first job? The employee whose workload doubles while pay stays flat? The artist whose style is absorbed into a model but whose loss never shows up in unemployment statistics? A targeted relief system would spend years fighting over proof. A universal cash floor skips the impossible detective work: the same transition that raises national wealth should raise household security.

Universalism also protects politics. Programs only for the visibly displaced become narrow, stigmatized, and easy to cut. A universal dividend builds a broad constituency that reads it as a share of public wealth, not a handout for failure.

The worst criticism of UBI in the AI debate is that it gives up on work. It gets things backwards. Basic income is how people keep agency when work changes. It is a platform for caring for others and taking risks.

In an AI economy this matters even more. If AI raises productivity, people should gain more freedom. If it compresses some jobs, people should gain more mobility. If it creates vast fortunes, the public that made it possible should gain more ownership — not just more advice to adapt.

The temptation will be to reach for the old, comfortable failures: paternalistic vouchers that tell people the state knows their lives better than they do; job training offered with no income behind it, which is just homework assigned during a crisis; or a narrow “AI-victim” bureaucracy that arrives too late and forces workers to prove harm in a market where harm often looks like an opportunity that never came. Unemployment insurance belongs on that list too — it was built for a clearer employment relationship than the AI economy is likely to provide.

And the case itself should not be rooted in the most extreme AGI scenario. That makes the policy look speculative when the real evidence is already visible in layoffs, hiring shifts, wage pressure, and concentrated ownership. We do not need a forensic unit to trace AI disruption back to its source. We need a universal floor.

The politics of AI and basic income are not as far apart as they look. The left can support an AI dividend because it redistributes concentrated technological gains and protects workers. The center can support it because it is simpler than expanding dozens of targeted programs that always lag reality. Conservatives can support it because it strengthens families, rewards public ownership, avoids paternalistic bureaucracy, and hands people cash rather than another managed-service maze.

The national-security case is stronger than it first appears. Societies under rapid technological stress become angry, distrustful, and vulnerable to demagogues. If people experience AI as a machine that enriches a few and destabilizes everyone else, they will turn against both the technology and the institutions that allowed it.

The bottom line

Basic income did not need AI to become morally serious. But AI has made the timeline harder to ignore. The responsible position is not panic; it is preparation. The labor market has not collapsed, yet employers are already citing AI for record job cuts. Recursive self-improvement has not arrived, yet a leading lab now says AI is increasingly building AI and that institutions may not be ready. The evidence does not show that the future is fixed. It shows that the range of plausible futures has widened, and the downside risks are now large enough to demand architecture.

A society that waits for perfect proof of disruption will build its support too late. By the time every dataset agrees, the bargaining power will have shifted, the entry-level jobs will have thinned, the wealth will have concentrated, and the public will again be told to accept insecurity as the cost of progress.

The better answer is direct and practical: build a universal cash floor, funded in part by the AI-era wealth that public knowledge and public infrastructure helped create.

AI may or may not become recursively self-improving soon, and the economy may or may not face mass technological unemployment. But the policy lesson is already clear.

If machines are going to learn faster, society has to protect people sooner.

Note on sources

Linked inline: BIEN’s basic income definition and its March 2026 report on the AI Dividend; the Challenger, Gray & Christmas May 2026 job-cuts report; Anthropic’s “When AI builds itself,” Economic Index, and labor-market-impacts research; Scientific American on the recursive-self-improvement warning; the U.S. Bureau of Labor Statistics May 2026 employment situation; Jones & Marinescu on the Alaska Permanent Fund Dividend (AEJ: Economic Policy, 2022); and the Stockton Economic Empowerment Demonstration.