How the OBR and the Bank of England Treat AI

by Claude Opus 5.5

How do the OBR and the Bank of England currently treat AI in their UK forecasts, and what evidence would make them change those assumptions?

Neither institution builds a separately quantified AI effect into its central forecast. Both treat AI as one reason productivity growth is expected to recover modestly, and both handle the bigger possibilities through scenarios and risks. The OBR has gone further on fiscal scenarios, setting out an AI-driven productivity upside, a technology-displacement case for unemployment and a warning about the tax base. The Bank has gone further on direct evidence, collecting firm survey data, Agents’ intelligence and staff analysis. What would change their central assumptions is measured productivity growth that persists, a sustained shift in jobs away from AI-exposed roles that shows up in unemployment and not just in vacancies, and a falling labour share.

The OBR

March 2026 Economic and Fiscal Outlook (3 March). In the central forecast, the OBR describes the gradual pick-up in total factor productivity (TFP) as reflecting “the fading impact of past negative shocks, the adoption of AI, and the boost from planning reforms”. TFP growth rises gradually to 0.8% by 2030, and medium-term productivity growth is about 1%. AI is therefore built in implicitly, but the OBR doesn’t attach a number to it. Around that central case it set out:

Productivity scenarios. In the upside, productivity growth reaches 1.5% a year, which “could, for example, be underpinned by a larger or faster than expected boost from AI”. Borrowing is then about £50bn lower in 2030–31. In the downside, growth stays at 0.5%, close to the post-financial-crisis average, and borrowing is about £40bn higher. The gap between the two cases is roughly £90bn of annual borrowing (computed).
A structural unemployment scenario (Box 2.2). Here “new technology displaces workers and is a substitute for labour”. Equilibrium unemployment rises to 5.5%, against 4.1% in the central forecast. Productivity rises for the workers who stay employed, and the level of GDP is broadly unchanged, but the productivity gain is “not reflected in higher real earnings”. The labour share falls, profits rise, and because labour income is taxed more heavily, the economy becomes less “tax-rich”. Secondary reporting put the cost at about £9bn a year in extra borrowing, but that figure does not appear in the OBR text reviewed for this answer.

July 2026 Fiscal Risks and Sustainability report (7 July). The long-term baseline assumes TFP growth driven by “the assumed impact of Brexit fading and a growing effect from AI”. The report says AI “has the potential to provide a significant boost to TFP as a general-purpose technology”. It notes that external estimates range from “up to 0.8 percentage points a year” down to “under 0.2 percentage points a year”. Its main points on AI:

A higher-productivity scenario. TFP grows at 1.3% a year, 0.3 points above the baseline. Debt reaches an unsustainable path almost 20 years later than in the baseline and is about 120 percentage points of GDP lower by 2075–76.
Box 4.1, “The potential impact of AI on the labour share and tax receipts”. The executive summary sets out the risk: if higher productivity “resulted partly from increased AI use”, it “could be accompanied by an AI-driven shift in the composition of GDP from (more highly taxed) labour to (lower-taxed) profits”. In other words, faster growth would not deliver all the fiscal gain the productivity scenario implies.

Over 2026, then, the OBR moved from mentioning AI to treating it in three separate ways: as a driver of growth, as a risk to employment and as a risk to the tax base. The productivity effect remains a scenario, not the central forecast.

The Bank of England

February 2026 Monetary Policy Report. Box C on the productivity outlook had potential productivity growth picking up “modestly” to “a little below its assumed long-run trend rate of 1%”. DMP firms expected AI to raise their productivity by about 0.6% a year over three years, and the Bank noted that UK businesses “may be some of the biggest adopters of AI so far”. It treated AI as a two-sided risk:

Stronger AI-driven productivity “could put temporary downward pressure on firms’ unit labour costs and require a looser policy stance, all else equal”.
If AI had already lifted potential productivity, “there may be more spare capacity in the economy”.
The Bank also noted that AI optimism has supported financial markets, and that a disappointment could reduce global and UK GDP.

The Agents reported firms planning to meet demand “by investing in automation and AI, rather than by raising headcount”.

July 2026 Monetary Policy Report (30 July). The language became more specific:

Firm expectations. DMP respondents “expected AI to reduce employment by around 0.4% per year and to boost productivity by around 0.9% per year over the next three years”. But “nearly 90% of DMP survey respondents report no material impact of AI on their employment over the past three years”.
Agents. They found “AI adoption is gradually reducing demand for highly automatable jobs in some industries”, with firms slowing hiring or leaving vacancies unfilled.
A new inflation channel. The rapid expansion of AI capacity is pushing up chip prices. AI-related demand is expected to raise UK-weighted world export prices by “a little over 1%” and to add “a little over 0.1 percentage points” to UK CPI inflation by the end of 2026.

Other Bank evidence in 2026:

Staff analysis (6 August). A Bank Underground post by Haley Schlicht built a combined exposure measure and found online job adverts down 15% in high-exposure occupations, 10% in mid-exposure and 6% in low-exposure ones. Customer-service adverts fell 23% and administrative adverts 22%. It concluded that “confident attribution remains premature”. A companion post, as reported, found software and IT consulting adding about 0.1 percentage points a year to productivity growth in 2023–25, ten times the pre-Covid rate. Bank Underground posts are staff views, not Bank policy.
Agents (September). They found “limited evidence of broad AI-driven reductions in employment”, with AI “influencing role design and replacement hiring” and “quantified productivity gains remain rare”.
Bailey’s speeches. On BBC Radio 4 (19 December 2025), Andrew Bailey said AI would “displace people” but probably not cause “mass unemployment”, and asked “what is it doing to the pipeline of people?” In Sheffield on 21 May 2026 (“Can AI make cutlery?”) he described AI as a likely general-purpose technology whose arrival “in the productivity numbers is an open question”. He said the employment outcome “depends on which of these forces dominates” (displacement, reinstatement and productivity) and that it “may well impact more skilled jobs, and so-called entry level jobs”.
Other work. Bloomberg reported in March that the Bank planned to war-game an “AI shock” for financial stability. In June the Bank’s Centre for Central Banking Studies hosted a conference on transformative AI and monetary policy.

The September 2026 MPC minutes mentioned AI only as a source of price pressure, through demand for AI-related technology goods, and not in connection with the labour market. The Committee’s attention was on inflation of 3.1%, projected to rise to around 3¾% in Q4 2026 and slightly above 4% in Q1 2027 on energy prices, and on weak labour demand.

Why both stay cautious

Three reasons:

The measured data don’t yet show a break. UK unemployment is 4.9%, vacancies are at their lowest outside Covid since 2014, and payrolls are down 145,000 on the year. All of this can be explained by weak demand and higher employment costs. Bloomberg Economics argued in June that AI is being wrongly blamed.
The productivity data are themselves uncertain. Reported LSE analysis puts annualised productivity growth at 1.6% between 2024 Q3 and 2026 Q1, against 0.3% in the previous decade. It stresses that the result depends on which employment series is used: payroll data show employment falling over the period, while the LFS shows it rising.
Getting it wrong is costly for both. An unjustified upgrade to the OBR’s productivity assumption would flatter the fiscal headroom. A premature AI story at the Bank could lead it to misjudge spare capacity at a time when the MPC is split on whether to raise rates.

What evidence would change their assumptions

For productivity (both institutions):

Several quarters of output per hour growing well above 1%, confirmed by both the LFS and payroll data, and concentrated in AI-using industries.
DMP firms reporting realised productivity gains, not just expected ones, and the Agents finding “quantified” gains becoming common.
A shift from AI as an operating cost (subscriptions) to capital investment, which would show firms redesigning processes.

For employment and equilibrium unemployment:

Falling adverts in exposed occupations turning into rising unemployment for those occupations or for young entrants, not just fewer hires.
A Beveridge-curve shift, meaning higher unemployment at a given vacancy rate, which would point to mismatch rather than weak demand.
Long-term unemployment rising further; the latest annual rise was already driven by people unemployed for more than six months.
A clearly rising share of DMP firms reporting realised AI headcount cuts, not just expected ones. At present nearly nine in ten report no effect.

For the tax base (OBR):

Labour’s share of national income falling while profits rise.
Income tax and National Insurance receipts growing more slowly than GDP, with corporation tax growing faster.
Real earnings lagging measured productivity, as in the OBR’s displacement scenario.

Timing: The OBR’s next forecast is due alongside the Budget, which is expected on 28 October 2026. The DMP and the Agents report every month and every quarter respectively. A plausible sequence is that the Bank revises its supply-side productivity assumption first, and the OBR follows by moving AI from a scenario into its central forecast.

Bottom line

As of October 2026 both institutions treat AI as a likely source of moderate productivity gains and a serious but unproven risk to jobs and taxes. Each has taken it seriously in scenarios and in collecting evidence, and neither has put a number on it in the central forecast. Neither will make a big change on firms’ expectations alone. They will need measured productivity, earnings and unemployment data to move first.

Sources

From AI and Jobs: UK, October 2026