--- title: "Assessment of AI capabilities and the impact on the UK labour market" aliases: - "DSIT UK Labour Market Assessment" - "DSIT 2026" - "UK Government AI Labour Assessment" tags: - uk - government - evidence-review - productivity - exposure - hiring - policy source: "/research/2026-01-dsit-uk-labour-market-assessment.pdf" extract: "/research/extracts/2026-01-dsit-uk-labour-market-assessment.md" authors: "DSIT AI and Future of Work Unit; AI Security Institute" publisher: "UK Department for Science, Innovation & Technology" date: "2026-01-28" --- # Assessment of AI capabilities and the impact on the UK labour market > [!info] Source > PDF (human): [2026-01-dsit-uk-labour-market-assessment.pdf](/research/2026-01-dsit-uk-labour-market-assessment.pdf) · Raw extract (machine): [2026-01-dsit-uk-labour-market-assessment.md](/research/extracts/2026-01-dsit-uk-labour-market-assessment.md) > DSIT + AI Security Institute — 28 January 2026 — Crown Copyright (Open Government Licence v3.0), 11 pages > URL: https://www.gov.uk/government/publications/assessment-of-ai-capabilities-and-the-impact-on-the-uk-labour-market ## TL;DR The UK Government's authoritative initial assessment of AI's labour-market implications. Five evidence judgements: 1. AI capabilities are improving rapidly in coding/cyber/research — task time horizon doubling every 7 months. 2. ~70% of UK workers are in occupations with AI-exposed tasks (vs ~60% in US/advanced-economy average) — UK has the highest exposure. 3. UK productivity gains could reach **0.4–1.2 percentage points annually** over the next decade (OECD), 2nd in G7. 4. Hiring is falling faster in AI-exposed occupations, but causality not established. 5. Significant evidence gaps remain — ongoing monitoring planned via the AI and Future of Work Unit. ## Key findings ### Judgement 1 — Capabilities improving rapidly in specific domains - Length and complexity of tasks autonomous AI agents can perform has approximately **doubled every seven months** in coding, cybersecurity, and research [METR]. - Anthropic data: AI agents are being used for **automation** more often than for assisting human work. - GDPval evaluation (1,300+ tasks from real professionals): best-performing model produced deliverables rated as good as or better than human expert output in **~50% of cases**. These tasks averaged 7 hours of expert time. ### Judgement 2 — UK is well-positioned but transition risks must be managed - **~70% of UK workers in AI-exposed occupations** vs ~60% in US/advanced-economy average — the **highest** in the IMF comparison. - UK exposure breakdown: 35% high-exposure & high-complementarity, 32% high-exposure & low-complementarity, 33% low-exposure. - US: 30% / 30% / 41%. - Advanced economy avg: 26% / 32% / 42%. - Roughly half of exposed UK workers are in **high-complementarity** roles (likely augmented). The other half are in **low-complementarity** roles (higher displacement risk). - OECD: UK labour productivity growth from AI could reach **0.4–1.2 percentage points annually** over the next decade — **2nd in G7** (only US higher). - UK labour productivity is currently **~20% below US levels**. - UK has ~23% of GDP in AI-exposed knowledge-intensive services. ### Judgement 3 — Productivity gains for specific occupations are real | Task | Productivity gain | |---|---:| | Writing tasks | 59% | | Software development | 56% | | IT support | 44% | | Legal work | 34% | | Consulting | 25% | | Commercial and R&D IT support | 20% | Note: these are speed-of-work changes, not output-per-employee. Cross-study methodologies differ. ### Judgement 4 — Hiring is falling in exposed occupations - UK job postings: 1 standard-deviation increase in AI exposure → **3.9% reduction in posting volume**. Effect statistically significant ~7 months after ChatGPT, intensifying thereafter; postings recover after ~20 months. - McKinsey: UK job adverts -38% for high-exposure occupations vs -21% for low-exposure (2022–2025). - US payroll study (Brynjolfsson): early-career employment in highly AI-exposed occupations -13%, stable or growing in less exposed. - **UK digital sector employment dropped for the first time in a decade in 2024**. **16–24-year-olds in computer programming -44% in a single year.** - Hospitality (low AI exposure) accounted for 53% of UK job losses Oct 2024–Aug 2025 — many job losses are unrelated to AI. ### Adoption (cited) - Business AI adoption has more than doubled since late 2023. - Only ~1 in 5 firms use or plan to use AI. - Within firms adopting AI, less than one-third of employees use it. - By size: 250+ employees: 36% adoption. 50–249: 23%. 5–9 (micro): 14%. - DSIT survey: 56% of using firms report productivity gains — most estimating up to 20% improvements (self-assessed). ### Counter-evidence cited - **Yale Budget Lab**: occupational mix is not changing faster than during previous technological transitions. Hiring declines in AI-exposed roles may be better explained by **2022 monetary policy tightening**, which disproportionately affected interest-rate-sensitive sectors (Information, Finance, Professional Services) that overlap with high AI exposure. - **Humlum & Vestergaard (Denmark)**: AI adoption had no measurable effect on worker earnings or hours, even for intensive daily users and early adopters. ### Judgement 5 — Significant evidence gaps remain - Causal links between AI adoption and employment/earnings. - Predicting which occupations/demographics face greatest risk or opportunity. - Measuring the rate of new job creation that may offset displacement. - Speed of worker transitions to new roles. - Task composition evolution within occupations. ## Implications for AdaptAI This is a **load-bearing source** — UK Government, recent, methodologically careful, comprehensive cited references. ### Calculator (`/calculator`) - **`industryModifier` — UK has higher exposure than US/avg.** Direction supported. - IMF complementarity split (35% high-comp / 32% low-comp / 33% low-exposure for UK) maps neatly onto the calculator's three-band exposure output (Low / Medium / High). Could be cited in the methodology modal. - **Time-horizon deltas:** the 0.4–1.2pp annual UK productivity gain is consistent with present caps (95/97/99 at 2/5/10 years) — supports current settings. - **Specific exposure data points:** - Software development +56% productivity - Writing +59% - IT support +44% - Legal +34% - Consulting +25% Map these to `automatedNow` / `automated2Years` for matching `jobTitlesWeighted` entries (Software Engineer, Marketing/Communications roles, IT support, Lawyer/Solicitor, Consultant). - **Age cohort:** -44% drop in 16–24-year-olds in computer programming is one of the most striking UK-specific data points in any of these papers — reinforces the case for an age-aware insight in the calculator output. ### Big Picture (`/big-picture`) - **`bigPictureData.keyStats` — concrete update candidates:** - Hero card "currentWorkforce" — could add "70% in AI-exposed occupations (vs 60% US avg)" as a UK-specific framing. - "projectedDecline" — refresh against OECD's 0.4–1.2pp productivity gain figure. - `bigPictureData.aiAdoptionSpeed.adoptionIndex` — UK firm adoption %s by size (36/23/14) are direct inputs. - `bigPictureData.industryExposure.aiExposure` — confirms order of exposure. - `bigPictureData.aiInvestment.investment` — DSIT figures useful. - `bigPictureData.unemploymentByAge.youth16to24` — the 44% drop in 16–24-year-old programmers is a load-bearing data point. - `bigPictureData.skillsDemand.{declining,growing}` — productivity gain numbers per task type are direct evidence. ### Industry Insights - `/industries/tech` — direct support: software development +56% productivity; programmer 16–24 employment -44%; UK digital sector total employment fell first time in a decade. - `/industries/finance` — high exposure with augmentation (high-complementarity in IMF framework). - `/industries/professional-services` — Legal +34%, Consulting +25% are direct citations. - `/industries/admin-customer-support` — DSIT survey supports admin/clerical decline framing. ### How to Adapt - `/adapt/upskilling` — DSIT's 5–9 employee firms 14% adoption gap identifies an SME-worker training need. - `/adapt/ai-tools` — the productivity gains by task type are exactly the right evidence to cite. - `/adapt/rights-protections` — references to the AI and Future of Work Unit are useful for the "what is government doing?" section. - `/adapt/career-transitions` — the IMF complementarity split (50/50 of exposed UK workers are high-comp / low-comp) is the right evidence for "your role may augment rather than replace". - `/adapt/human-advantage` — AI agents are used for **automation more often than assisting** [Anthropic source] — useful counterweight to vendor "augmentation only" claims. ## Related notes - [[CBP UK Evidence Review]] — independent UK evidence review covering similar territory with more granular adoption data. - [[GLA Working Paper 103]] — London-specific application of the same exposure framework, with sectoral breakdown. - [[BIS European Firm-Level Study]] — supports the productivity gain figures cited in Judgement 3. - [[IMF Skill Gaps SDN]] — IMF complementarity framework explicitly used in Judgement 2. ## Caveats & limitations - Government produces this as an "initial high-level assessment"; explicitly calls out evidence gaps. - Exposure indices have not been extensively validated against real-world outcomes. - 50%-of-tasks-rated-as-good-as-experts (GDPval) covers precisely-specified, self-contained digital tasks — real work involves greater ambiguity. - Productivity self-reports are not objectively measured. - McKinsey job-ad numbers are correlational, not causal. - The 7-month doubling is in coding/cybersecurity/research — uncertain if it generalises. - 56% of using firms reporting gains is self-assessed; limited robust statistical evidence at firm-level.