---
title: "London's workforce exposure to generative artificial intelligence"
aliases:
  - "GLA Working Paper 103"
  - "London Workforce GenAI Exposure"
  - "Dwan-O'Reilly 2026"
tags:
  - uk
  - london
  - government
  - exposure
  - ilo-framework
  - age-gradient
  - working-paper
source: "/research/2026-04-gla-london-workforce-genai-exposure.pdf"
extract: "/research/extracts/2026-04-gla-london-workforce-genai-exposure.md"
authors: "Jeff Dwan-O'Reilly"
publisher: "GLA Economics — Greater London Authority"
date: "2026-04"
---

# London's workforce exposure to generative artificial intelligence

> [!info] Source
> PDF (human): [2026-04-gla-london-workforce-genai-exposure.pdf](/research/2026-04-gla-london-workforce-genai-exposure.pdf) · Raw extract (machine): [2026-04-gla-london-workforce-genai-exposure.md](/research/extracts/2026-04-gla-london-workforce-genai-exposure.md)
> GLA Economics — Working Paper 103 — April 2026 — Regional government economic research (UK / London-specific), 73 pages
> Foreword by Sir Sadiq Khan, Mayor of London

## TL;DR

A rigorous London-specific exposure analysis using the ILO's 2025 task-based GenAI framework adapted to UK SOC 2020. **At least 46% of London's workforce — about 2.4 million people — are in roles where GenAI could automate a share of their tasks**, substantially higher than the UK average of 38%. ICT, finance, professional services, and public administration sectors face the highest exposure. Over **300,000 routine administrative workers face the highest levels of exposure and risk of automation**. Worker-/business-reported AI adoption has roughly doubled in two years to 26–35%. Mayor Khan announces a London AI and Jobs Taskforce.

## Key findings

### Exposure scale

- **46% of London workers (~2.4M)** in roles where GenAI could automate a share of tasks. UK average: 38%.
- **>300,000 workers in highest-exposure routine administrative roles.**
- IMF prediction (cited): ~70% of jobs in the UK could be impacted to some extent within five years.

### Sectoral and occupational exposure

- **Highest exposure sectors (by occupational mix):** ICT, finance, professional services, public administration.
- **Lower exposure:** sectors relying on physical presence, skilled trades, direct interpersonal care.
- Many professional white-collar occupations are highly exposed (language-, digital-, analysis-intensive). Roles requiring high-stakes judgement, accountability, client interaction less directly automatable — more likely augmented than replaced.
- Note: low-exposure sectors like transport may face faster change as **computer vision and robotics** advance (separate from GenAI).

### Distributional patterns

- **Higher-educated workers more exposed** (concentrated in professional, knowledge-intensive occupations).
- **Women overrepresented in highly exposed administrative/clerical roles.**
- **Younger workers concentrated in high-exposure digital and knowledge-intensive roles.** Particular concern for "stepping-stone" jobs that traditionally provide entry routes into careers.
- Potential to **widen income inequality**: greater disruption in lower-paid admin roles, productivity gains accruing to higher-paid professionals or model owners.

### Adoption

- Worker-/business-reported AI adoption has **almost doubled over the last two years to between 26%–35%**.
- Adoption uneven — larger and more digitally mature businesses lead. Concerns around data protection, quality, accountability slow adoption in some settings.

### Observed effects so far

- Task change within roles is the main employment effect to date.
- Sub-task automation and role redesigns rather than wholesale job replacement.
- Firms prioritise training and upskilling existing staff in complementary skills.
- Increasingly expect AI expertise when recruiting across a wide range of occupations.
- Demand for **specialist AI roles is growing rapidly** but remains a relatively small share of the labour market.
- Early signs of **slowing recruitment for some most-exposed admin and professional roles**, but evidence inconclusive.
- Limited reported AI-related headcount reductions to date; some employers anticipate efficiency-driven reductions over time.

### Three sectors prioritised by the new Taskforce

1. **Professional and financial services** — economic importance + concentration of highly AI-exposed roles.
2. **Creative industries** — both commercial creative (marketing, design) and culturally creative (media, arts production).
3. **Transport** — limited GenAI impact today, but more disruptive automation effects plausible medium-term.

### Channels through which GenAI affects labour demand (OECD framework adopted)

1. **New task creation** (technology generates new labour demand).
2. **Task automation** (substitution and potential displacement risk).
3. **Task augmentation** (human-technology complementarity transforms role).

### Factors shaping outcomes (Table 1.1 in paper)

- Worker-occupation: degree of task–technology overlap, worker trust in AI, scope for task reorganisation, complementary skills/AI proficiency.
- Business: adoption/integration intensity, strategic objective (cost cut vs growth), organisational capability, market competition.
- Wider: public trust, security/regulatory compliance, macroeconomic conditions, demand responsiveness, collective bargaining, infrastructure capacity.

## Methodology in brief

Adapts the **International Labour Organization's 2025 task-based GenAI exposure framework** (Gmyrek et al. ILO, 2025) to London via ISCO-08 → SOC 2020 crosswalks. Starts from ~30,000 tasks rated for susceptibility to GenAI capabilities by worker judgements + expert review, then scaled by two LLMs. Aggregates to occupations using mean exposure and dispersion. Higher and more uniform exposure interpreted as more automation-prone; lower or more variable exposure as more augmentation-prone.

## Implications for AdaptAI

This is the **single most relevant UK-specific source** for AdaptAI in this batch — London-specific and rigorous, with explicit demographic and sectoral breakdowns.

### Calculator (`/calculator`)
- **`industryModifier`:** ICT (+10), Finance (+9), Professional Services (+8), Public Admin (+4) — directional support. Public Admin should consider being raised given the explicit identification of public administration as one of the four highest-exposure sectors.
- **Headline framing:** the Mayor's foreword line "high exposure does not signal definitive job losses, nor does low exposure guarantee security" matches AdaptAI's existing tone (calculator presents probability, not destiny).
- **`emotionalIntelligence` weighting:** roles requiring high-stakes judgement, accountability, client interaction are the augmentation-prone bucket. Validates the strong protective weight (-20 at level 4).
- **`physicalDexterity` weighting:** sectors relying on physical presence/skilled trades less exposed — supports current protective weight. **But flag** the paper's note on robotics/computer vision: this is the right place to start watching for a sign-flip on physical-dexterity protection in a few years.
- **Age-aware insight:** paper specifically calls out younger workers in stepping-stone jobs — the calculator already has answer-driven insights and could add an age cohort question if appropriate.

### Big Picture (`/big-picture`)
- **`bigPictureData.regionalData`:**
  - **London & SE highExposureJobs is currently 42** in the dataset. Paper's 46% figure for London alone is consistent — London-only would push the SE-weighted figure even higher. Worth a refresh.
  - `regionalData.avgAIInvestment` for London (currently £285/worker) — paper does not give a direct figure but supports the pattern of London being the leading adopter.
  - `regionalData.projectedImpact` (-5.2% for London by 2030) — paper does not directly forecast a London-specific 2030 figure, but the framing "potential to widen income inequality" is consistent.
- `bigPictureData.industryExposure` — supports current ranking; ICT/Finance/Professional Services/Public Admin as top tier.
- `bigPictureData.aiAdoptionSpeed` — 26–35% adoption almost doubling in 2 years is direct input for `workforceUsingAI`.

### Industry Insights
- `/industries/tech` — supports high exposure framing.
- `/industries/finance` — supports high-exposure-with-some-augmentation framing.
- `/industries/professional-services` — direct support; one of three Taskforce priorities.
- `/industries/public-sector` — explicit identification as a top-4 exposure sector. **Currently has no chart data in `industryEmploymentData` — this paper provides the rationale and direction to add it.**
- `/industries/creative` and `/industries/creative-media` — Taskforce priority sector; refresh prose.
- `/industries/admin-support` and `/industries/admin-customer-support` — 300,000+ Londoners in highest-exposure admin roles is a direct citation.
- `/industries/logistics` — Taskforce mentions transport as a sector facing future disruption from non-GenAI AI (computer vision, robotics).

### How to Adapt
- `/adapt/upskilling` — paper explicitly supports training/upskilling in complementary skills as the dominant firm strategy. Cite firms increasingly expect AI expertise when recruiting.
- `/adapt/ai-tools` — observed effects to date are sub-task automation + role redesign, not wholesale replacement. Supports advice to learn AI tools as augmentation.
- `/adapt/repetitive-tasks` — 300,000 London admin workers at front of queue is exactly the audience.
- `/adapt/human-advantage` — high-stakes judgement, accountability, client interaction as the protected bucket is precisely the framing.
- `/adapt/rights-protections` — Mayor's foreword + Taskforce announcement give a UK-relevant policy hook (workers' representation, employer responsibility, training infrastructure).
- `/adapt/career-transitions` — stepping-stone jobs hollowing out is the right frame for early-career advice.

## Related notes

- [[DSIT UK Labour Market Assessment]] — UK national-level government complement, citing the same IMF/ILO frameworks.
- [[CBP UK Evidence Review]] — UK national-level evidence review with adoption + employment outcomes.
- [[IMF Skill Gaps SDN]] — IMF cross-country skill-demand framework that informs the London exposure framing.

## Caveats & limitations

- Exposure measures the **technical potential** for tasks within roles to be automated by GenAI. **Not** a forecast of jobs gained or lost.
- Mayor's foreword explicitly notes high exposure doesn't equal job losses; low exposure doesn't guarantee security.
- ILO methodology dated to early 2025; AI capability advancing fast — benchmarks become outdated quickly.
- Adoption (26–35%) is self-reported by workers/businesses.
- Hiring slowdown evidence is "tentative" / "still inconclusive" — paper cautions against over-interpreting.
- ISCO-08 → SOC 2020 crosswalks are imperfect.
- Methodology covers GenAI specifically — does not include robotics, computer vision, or non-generative AI applications.
- Snapshot analysis; the report itself frames as a "foundation for further work" by the new Taskforce.