The political consensus on UK welfare is to tighten. The Universal Credit Act has been amended. The Timms Review of Personal Independence Payment is in flight. Eligibility is narrowing. The argument behind the tightening is a working-market argument: people of working age can work, the safety net should incentivise that work, and the cost has grown too far. That argument assumes labour markets stay roughly intact. Reading the receipts, that assumption is already wrong before generative AI displacement fully lands. This article reads three sources and asks what the math says.

This is not a left-versus-right framing. The first half of the OBR data alone makes the case that the UK welfare system, on its current track, stops working as a safety net within five years. What replaces it is the open question. Universal Basic Income is one of three viable answers, alongside an expanded Universal Credit plus retraining grants, or further conditionality and tightening that pushes more people off support entirely. The cut-cut-cut consensus picks the third path. The OBR's own caseload modelling says the third path likely does not hold.

UK incapacity benefit spending and growth: the OBR numbers

The Office for Budget Responsibility published its latest Welfare Trends Report in October 2024. Working-age incapacity benefits (Employment and Support Allowance and the Universal Credit health element) cost the Treasury £24.9 billion in 2023-24. Working-age disability benefits (Personal Independence Payment and Disability Living Allowance) cost a further £19.0 billion. Together, more than £43 billion goes through these two channels alone before the State Pension, Universal Credit standard allowance, and Housing Benefit are counted.

The trend matters more than the level. Incapacity benefits prevalence as a share of the working-age population was 5.4% in 2019-20. It was 7.0% by 2023-24. The OBR projects 7.9% by 2028-29, which it describes as an all-time high. The OBR's own commentary is that the incapacity benefits system has ‘rarely reached any kind of steady state’ over the past half-century. Whatever steady state means, the trajectory from 5.4 to 7.9 in less than a decade is not it.

That growth happened mostly between 2019 and 2024. Whatever drove it, generative AI did not. ChatGPT launched in late 2022. The first business-meaningful generative-AI deployments in UK firms are still being rolled out. The pressure showing up in the OBR caseload is pre-AI. It reflects long-COVID effects on the labour market, mental-health prevalence rising in the post-pandemic working-age cohort, and the operational consequences of moving people on to Universal Credit. None of that goes away when AI displacement starts.

How many UK jobs are at risk from AI automation?

The Office for National Statistics published an automation-risk analysis in March 2019. It examined 20 million jobs in England based on 2017 data. It concluded that roughly 1.5 million jobs were at high risk of automation, representing 7.4% of the roles analysed. That figure had ticked down slightly from 8.1% in 2011. The highest-risk roles named in the analysis were waiters and waitresses, shelf fillers, and elementary sales occupations.

The analysis is dated by design. It cannot account for capabilities that did not exist when it was written. ChatGPT and the wave of generative-AI products that followed have expanded the set of cognitive tasks that machines can perform, not the set of physical tasks. The 2019 ONS list was heavy on physical-routine roles. The next list, when it is written, will read differently. Roles like junior copywriting, mid-tier paralegal review, customer-service first-line response, and entry-level coding work were not on the 2019 high-risk list. They will be on the next one.

That redistribution of risk is what changes the welfare math. The 2019 analysis told a story in which automation risk was concentrated in occupations whose workers had less human capital and fewer transferable skills. The post-2022 analysis tells a story in which automation risk also touches occupations whose workers have degrees, professional qualifications, and the kinds of mid-life mortgages and dependants that make occupational disruption financially serious.

Does Universal Basic Income reduce work? What three pilots found

Three large empirical Universal Basic Income pilots have produced peer-reviewed results that are worth reading before forming a position on whether UBI ‘works’.

Finland UBI pilot (Kela, 2017-2018) results

The Finnish social insurance institution Kela ran a randomised pilot of 2,000 people aged 25 to 58 who were drawn from those receiving unemployment benefits. Each participant received an unconditional payment of €560 per month for two years, retained even if they found work. The headline finding on the most-asked question was that employment levels did not change between the treatment and control groups. The headline finding on wellbeing was that the treatment group reported fewer stress symptoms, fewer difficulties concentrating, and fewer health problems, and was more confident about the future.

The Kela pilot is the strongest empirical answer to the most common objection to UBI: that unconditional cash reduces work. Across two years, with 2,000 randomly selected unemployment-benefit recipients, it did not.

Stockton SEED Universal Basic Income pilot results

The Stockton Economic Empowerment Demonstration enrolled 125 residents living at or below the city's median income. Each received $500 per month for eight months. The final report found participants spent the payments primarily on groceries and bills. Around 43% of participants held a full-time or part-time job during the pilot. SEED was a smaller pilot, more limited in duration than Kela, but it answered the same employment question with the same direction: cash did not push recipients out of work.

Kenya GiveDirectly Universal Basic Income study results

The GiveDirectly long-term study covers 20,000 recipients in 195 rural villages and runs over either two-year or twelve-year arms depending on the study group. It is the largest UBI study currently active anywhere. Reported findings include lower food insecurity, better physical and mental health, and higher rates of business creation among recipients than among comparison groups. The Kenyan context is different from the UK in obvious ways. The relevant transfer is the methodological evidence that unconditional cash, in a randomised design, does not produce the dependency outcomes that the cut-cut-cut framing predicts.

UBI vs UK welfare reform: three policy paths

Combine the three readings. The OBR caseload data shows working-age incapacity prevalence rising from 5.4 to 7.9 in less than a decade, on track to an all-time high in 2028-29, before the AI-displacement layer fully lands. The ONS pre-AI baseline put 1.5 million jobs at high risk of automation, with the next analysis likely to redistribute risk into cognitive and white-collar occupations. The three large UBI pilots show that unconditional cash does not cut employment but does measurably reduce financial and health stress.

Three policy paths follow. The current government has chosen the third by default: tighten eligibility, narrow descriptors, push more working-age people off support. That is a bet that the labour market absorbs everyone the welfare system is removing. The OBR caseload data is the pre-AI evidence that the labour market is not absorbing them now. Layering generative-AI displacement on top of that does not improve the bet.

The first path is to expand Universal Credit substantially and pair it with conditional retraining grants. That keeps the work-incentive logic of the existing system but funds the transitions the AI-displaced will need. It is more honest than tightening but more expensive than the current envelope.

The second path is Universal Basic Income at a level that pays the floor of subsistence. The pilot evidence above is the case that the dependency objection does not survive contact with the data. The case against UBI is fiscal: at a level that meaningfully replaces working-age welfare, the gross cost is large. The case for UBI is that it removes the assessment regime, the descriptor scoring, the appeals tribunal pipeline, and the fluctuating-condition argument from the welfare bureaucracy entirely. PIP cost £19.0 billion in 2023-24 and is rising.

The third path is the current policy. Tighten until labour markets absorb the displaced. The OBR data and the pilot evidence say it does not.

What this means for UK welfare reform 2026 and beyond

The UK welfare debate has become tribal because both sides argue ideology. The receipts argue something simpler. The system is already under structural strain. AI displacement, even on cautious estimates, does not improve that. The current tightening is a response to existing pressure, not to the bigger pressure that is coming. UBI is one of three viable answers and the only one that does not require the assessment-and-appeal apparatus that produces the cases the Timms Review now exists to scrutinise.

The political question is which of the three answers gets picked. The empirical question is whether the third answer (the current path) holds. The OBR has spent the past decade publishing data that says it does not.

Frequently asked questions

Is the UK welfare system actually under strain right now?

The Office for Budget Responsibility’s October 2024 Welfare Trends Report shows working-age incapacity benefits cost £24.9 billion in 2023-24 and disability benefits cost a further £19.0 billion. Incapacity benefits prevalence as a share of the working-age population was 5.4% in 2019-20, was 7.0% in 2023-24, and is projected at 7.9% by 2028-29 — an all-time high.

How many UK jobs are at risk from AI automation?

The most recent ONS analysis, published in March 2019, found around 1.5 million UK jobs at high risk of automation, equivalent to 7.4% of the 20 million jobs analysed. That figure pre-dates the 2022-onwards generative AI deployment cycle, which has materially expanded the set of cognitive tasks AI systems can perform. The next analysis is likely to redistribute risk into white-collar and cognitive occupations that were not on the 2019 high-risk list.

What did the Finland Universal Basic Income pilot actually find?

The Finland Kela pilot ran from 2017 to 2018 with 2,000 randomly selected unemployment-benefit recipients aged 25 to 58. Each received an unconditional €560 per month for two years. Employment levels did not change between the treatment and control groups. The treatment group reported fewer stress symptoms, fewer difficulties concentrating, and fewer health problems, and was more confident about the future.

Would Universal Basic Income reduce work incentives in the UK?

The Finland Kela randomised trial, the Stockton SEED demonstration (125 participants, $500 per month for eight months, 43% of participants in work), and the Kenya GiveDirectly long-term study (20,000 recipients across two-year and twelve-year arms) all found unconditional cash transfers did not reduce employment. The empirical evidence cuts against the dependency objection.

How much does the UK currently spend on working-age welfare?

Per the OBR October 2024 Welfare Trends Report, working-age incapacity benefits cost £24.9 billion and working-age disability benefits cost £19.0 billion in 2023-24. Together that is more than £43 billion through these two channels alone, before the State Pension, the Universal Credit standard allowance, and Housing Benefit are counted.