Likelihood, two years: 3Possible; to end-2028
Likelihood, ten years: 4Likely; to end-2036
Systemic impact: 3Significant; global
National impact: 4Severe; typical highly exposed nation
Onset pace: Months
Duration (acute phase): 3 years
Warning time: Months
Scope: Global
Recovery horizon: Structural
Capability load: Hard 1/3Soft 3/3Economic 2/3; domains loaded High
Concurrency: Standalone; triggers 3 · triggered by 4
Confidence · movement: low; new

Rated at Standard Severe. Event reach: global event (The rating is for displacement at the stated share and period. Gradual automation of tasks is a trend and is not rated here.). Source of scores: model-panel-v0.5.

Load test: results by nation

03Narrative

Dateline: March 2031

The step came in the first half of 2028. Within a few months, three providers released agent systems that completed administrative and professional tasks lasting several days, at an error rate below that of a supervised junior employee and at a cost per task that fell by more than half during the year. Until then, firms had used AI to assist staff. From the autumn of 2028 they assigned it whole workflows: claims handling, accounts payable, contract review, first-line customer service, software maintenance and routine reporting.

Headcount did not fall first. Recruitment did. In the highly exposed economies, graduate and trainee intakes in the affected occupations for 2029 were about half those of 2027. Redundancies followed through 2029 and 2030, as firms redesigned processes around the agents and as slower competitors lost contracts to faster ones. By early 2031, jobs equal to 10 per cent of employment in these economies had been removed or changed so far that the previous holder could not fill them without retraining. About a third of the displaced workers had found work at comparable pay.

Output rose. Measured productivity in business services grew faster than at any time in two decades, and corporate profits rose with it. The gains were concentrated in the firms that owned the models and in the early adopters. Public revenue moved the other way. By the third year, income tax and social contributions were 10 per cent below budget plans, because those plans had assumed that rising output would bring rising payrolls. Corporate tax receipts increased by less than the payroll shortfall, and part of the profit was booked abroad.

Benefit systems designed for cyclical unemployment met claimants who were qualified, in mid-career and unlikely to be recalled. Public employment services had training places for a small fraction of them, and many courses prepared people for occupations that were themselves contracting. Universities recorded falling applications for law, accountancy and computing degrees. Professional bodies reported that the supervised junior work through which practitioners had qualified no longer existed in sufficient volume.

Services exporters were affected through trade. Clients renegotiated outsourcing contracts for back-office work and software maintenance at lower volumes or took the work back in-house. Export receipts from business services fell in several South Asian and South-East Asian economies, where the sector had been the main source of graduate employment.

Surveys in 2030 recorded a marked fall in the share of young adults who expected to be better off than their parents. Legislatures debated taxes on automated work, restrictions on deployment in regulated professions and wage insurance. Few had legislated by 2031. The nations that held their position had kept entry routes open through funded traineeships, moved the tax base towards profits and consumption, retrained at scale and used the same systems to improve public administration.

The dateline is illustrative, not a forecast. The narrative is hypothetical; the historical anchors below are real events.

04Summary

Reliable, low-cost AI agents make complete administrative and professional workflows commercially viable within a period of months. Over the following three years, employers in highly exposed economies remove or radically redesign jobs equal to 10 per cent of employment. Entry-level recruitment in affected occupations halves before training and benefit systems adapt. Productivity gains accrue mainly to model owners and leading adopters, while employment-linked tax receipts fall 10 per cent below plan. Services exporters lose outsourcing contracts. The test is whether a nation can absorb the displacement without losing its supply of skilled entrants, its public revenue or public trust in economic opportunity.

05Historical anchors

Event	Date	What happened	Calibrates
Generative AI adoption	2023–2026	ChatGPT was reported to have reached about 100 million monthly users within two months of its release in November 2022. A Stanford Digital Economy Lab study of US payroll records, published in August 2025, found a 13 per cent relative fall in employment among workers aged 22 to 25 in the most AI-exposed occupations since late 2022, while employment of older workers in the same occupations held.	The order of events. Displacement appears first as lower entry-level hiring and not as redundancy, and it is measurable before whole workflows are automated. The scenario scales that early effect up.
METR time-horizon measurements	March 2025 to April 2026	METR measured the length of software tasks that AI agents complete with 50 per cent reliability and reported in March 2025 that it had doubled about every seven months since 2019.	The deployment step. If the trend holds, agents reach tasks lasting days within the two-year horizon. The measurements cover software tasks at 50 per cent reliability, which is well short of the reliability that unsupervised commercial use requires.
China after WTO accession	2001–2010	Research by Autor, Dorn, Hanson and co-authors attributed about one million US manufacturing job losses between 1999 and 2011 to import competition from China and about 2.4 million once linked industries and local demand are counted. Affected local labour markets showed depressed employment and wages for more than a decade.	Adjustment speed and persistence. A concentrated displacement much smaller than the one assumed here was not absorbed by mobility or retraining within ten years, and it had lasting political effects in the affected areas.
Global financial crisis	2008–2009	US payroll employment fell by about 8.7 million between early 2008 and early 2010 and the unemployment rate reached 10 per cent in October 2009. US federal receipts fell by about 17 per cent in fiscal year 2009. Studies of graduates who enter the labour market in a recession find earnings losses that persist for about ten years.	The scale of the employment and revenue loss and the lasting cost to cohorts that cannot enter. The difference is that in 2008 output fell and later recovered. Here output rises and the jobs do not return.
TCS workforce reduction	July 2025	Tata Consultancy Services, India's largest IT services exporter, announced that it would reduce its workforce by about 2 per cent, roughly 12,000 roles, mostly in middle and senior grades. The company cited skill mismatch and limited scope for redeployment.	The trade channel. Services exporters adjust headcount when the delivery model changes, before any fall in client demand. The scenario assumes a reduction several times larger across the sector.
Previous general-purpose technologies	1990s–2000s	Business computing spread through the 1970s and 1980s without a visible effect on measured productivity, a gap that Robert Solow noted in 1987. US productivity growth rose from the mid-1990s. Electrification of factories earlier took several decades to change output per worker.	A counter-anchor on speed. Firms have needed many years to reorganise around a new technology. This is the main reason the two-year likelihood is held at Possible.

06Parameters

Shown at their preset values. Parameters are not adjustable in this release and nothing on this page is computed from them. Custom settings run (Phase B) but are labelled 'non-standard run' and excluded from comparisons.

Common variables at Standard Severe · read-only

1. Severity: major
severe (Standard Severe)
extreme
2. Duration (acute phase): 30 days
90 days
1 year
3 years (Standard Severe)
5 years
3. Onset pace: days
weeks
months (Standard Severe)
years
4. Warning time: none
days
months (Standard Severe)
5. Scope: national
regional
global (Standard Severe)
6. Origin: natural
accidental (Standard Severe)
adversarial (great power / neighbour / non-state)
negotiated
Standard Severe: accidental (technological)
7. External support: full
partial
none (Standard Severe)
8. Concurrency: standalone (Standard Severe)
plus one named scenario
plus two
9. Policy response assumed: none (pure exposure)
current plans executed (Standard Severe)
best practice
10. Recovery horizon: months
years
structural (Standard Severe)

Scenario-specific parameters · read-only

Parameter	Default	Range or options	Note
Jobs removed or radically redesigned in highly exposed economies	10 per cent	5 to 20	Major 5, Severe 10, Extreme 20. A redesigned job is one the previous holder cannot fill without retraining.
Period of displacement	3 years	2 to 5	Measured from the point at which whole workflows become commercially viable.
Fall in entry-level recruitment in affected occupations	50 per cent	30 to 70	Arrives in the first recruitment cycle, before redundancies.
Shortfall in employment-linked tax receipts	10 per cent	5 to 15	Income tax and social contributions. Gains in corporate and consumption taxes are not netted off.
Displaced workers re-employed at comparable pay within two years	one third	one fifth to one half	An assumption. It sets benefit demand and the scale of retraining required.
Fall in business services export receipts for outsourcing economies	20 per cent	10 to 40	An assumption. Applies to nations where computer and business services are a leading export.
Domestic frontier-model capability	none	options: none, some, frontier	Determines whether the productivity gain and the taxable profit arise at home or abroad.

07Transmission channels

1
Agent reliability and cost pass the point at which firms can assign whole administrative and professional workflows, not single tasks.
2
Employers in exposed occupations halve graduate and trainee intakes in the first recruitment cycle.
3
Firms redesign processes around agents. Redundancies follow, and slower firms lose contracts to early adopters.
4
Clients renegotiate or withdraw outsourcing contracts, reducing business services exports and graduate employment in exporting economies.
5
Income tax and social contributions fall 10 per cent below plan while output and profits rise, partly outside the taxing jurisdiction.
6
Benefit and retraining systems built for cyclical unemployment meet qualified claimants who will not be recalled.
7
Applications for exposed degrees fall and the supervised junior work that qualifies professionals becomes scarce.
8
Expectations of mobility fall among young adults. Demands for taxes and restrictions on automated work divide politics and affect investment.

08Capability loading

High: capability band shifts expected under current plans. Medium: band shifts under 'none' policy response only. Low: strain without band shift. Loads are judgement-based until the Atlas connects. Domains link to the Atlas.

Domain	Load	Channel
Hard
Defence and security	Low	Civilian defence staff and contractors are affected; reliance on foreign models is limited to support functions.
Strategic infrastructure	Medium	Data centre, grid and connectivity capacity determine who can deploy agents at scale.
Critical technology	High	Access to frontier models and compute decides who earns the productivity gain and who pays for it.
Soft
Government effectiveness	High	Tax design, benefit administration, employment services and regulation of automated work are tested together.
Human capital	High	Entry-level routes close; retraining capacity and graduate skills fall behind the change in tasks.
Influence and cohesion	High	Lower expected mobility among young and qualified workers weakens trust and sharpens distributional conflict.
Economic
Macro-financial	High	Employment-linked receipts fall 10 per cent below plan, the labour share of income falls and benefit spending rises.
Industry, trade and supply	High	Business services exporters lose contracts; adopting firms gain share and market concentration rises.
Energy and resources	Medium	Inference demand raises electricity load where data centres cluster.

09Stakeholders

Government: Exposure
Receipts from income tax and social contributions, benefit and retraining systems, the public sector's own administrative workforce and the regulation of automated work in licensed professions.
Actions
Model revenue by tax line on a 10 per cent displacement over three years with output rising
Fund traineeships and wage insurance that keep entry routes open in exposed occupations
Set the capacity of retraining and employment services against the displacement assumed, not against a cyclical downturn
Watch
Monthly receipts from income tax and social contributions against forecast
Graduate and trainee intakes at the largest employers in exposed occupations
Benefit claims from qualified workers in mid-career
Technology: Exposure
Model owners and early adopters gain revenue and market share. They also face liability for agent errors, regulation of deployment and proposals to tax automated work.
Actions
Record agent error rates and human oversight by workflow before removing roles
Keep a supply of trained junior staff for the supervisory work that agents require
Prepare for deployment rules and reporting duties in regulated professions
Watch
Agent reliability on tasks lasting days and the cost per completed task
Liability rulings on errors made by automated systems
Legislative proposals on taxing or restricting automated work
Investors: Exposure
Labour-intensive business services, outsourcing firms, private higher education, household credit held by displaced professionals and sovereigns that depend on labour taxation.
Actions
Test portfolios on a 10 per cent employment displacement with output rising and labour income falling
Review sovereign exposures where income tax and social contributions provide most revenue
Assess household credit books by borrower occupation
Watch
Revenue per employee and headcount guidance at business services firms
Arrears on mortgages and consumer credit among professional households
Fiscal forecasts revised for lower payroll growth
Public: Exposure
Employment and earnings in administrative and professional occupations, the value of degrees that led to them and the first jobs through which young people enter skilled work.
Actions
Check which tasks in an occupation are already assigned to AI agents before choosing a course or career
Use retraining entitlements early, while employed
Ask employers and professional bodies how new entrants will be trained
Watch
Entry-level vacancies in the chosen occupation
Graduate employment and earnings by degree subject
Changes to benefit and retraining entitlements

10Regional exposure

Region	Exposure	Rationale
North America	High	Home of the leading model owners and early adopters. Large professional services employment and flexible dismissal rules bring fast displacement.
Europe	High	High shares of administrative and professional employment, revenue that depends on labour taxation and few domestic frontier providers.
China	High	Domestic models and rapid adoption, with graduate unemployment already high and a large clerical workforce.
Indo-Pacific	High	Outsourcing in the Philippines and services centres in Singapore and Australia are exposed. Ageing workforces in Japan and Korea soften the effect.
South Asia	High	IT and business process exports are the main source of graduate employment in India and a growing one in its neighbours.
Gulf and Middle East	Medium	Public sector employment and expatriate professional labour are exposed. Capital for adoption is available and labour taxation is light.
Africa	Medium	Small formal services sectors limit direct displacement. Outsourcing centres and the route to services-led growth are at risk.
Latin America and Caribbean	Medium	Nearshore services and urban professional employment are exposed. Large informal sectors limit the measured share.
Russia and Eurasia	Medium	Restricted access to frontier models slows adoption. IT outsourcing and public administration are exposed.

11Early-warning indicators

Indicator	Source	Threshold
Employment of workers aged 22 to 25 in AI-exposed occupations, relative to less exposed occupations	National labour force surveys; ILOSTAT	Relative fall of more than 20 per cent from the 2022 level
Job postings for entry-level roles in administrative, legal, finance and software occupations	Indeed Hiring Lab	Down 30 per cent on the 2024 level for two consecutive quarters
Length of tasks that AI agents complete reliably	METR	Time horizon at 50 per cent reliability above one working week
Price of model inference at a fixed level of capability	Stanford AI Index	—
Share of firms using AI to produce goods or services	US Census Bureau Business Trends and Outlook Survey; Eurostat	Above 30 per cent of firms in business services
Receipts from income tax and social contributions against budget forecast	National finance ministries; OECD Revenue Statistics	Shortfall above 3 per cent in a year when output grows
Labour share of national income	OECD; ILO	Fall of two percentage points within three years
Exports of computer and other business services from major outsourcing economies	WTO and UNCTAD trade in services statistics	Year-on-year fall in two consecutive quarters

12Compounds

Triggers: S05 Legitimacy rupture · Loss of expected mobility among young and qualified workers raises the stakes of a contested transfer of power.
S02 Global funding seizure · Revenue shortfalls, arrears among professional households and a repricing of labour-intensive services firms weaken sovereign and credit positions.
S17 Automated discovery step · The same step in agent reliability, applied to research
Triggered by: S17 Automated discovery step · The capability that lets AI systems direct research also completes administrative and professional workflows.
S13 Great-power détente · Relaxed technology restrictions widen access to frontier models and chips and speed deployment.
S15 Clean power threshold · Cheaper firm power lowers the cost of inference at scale.
S18 Public service leap · Automated case handling removes clerical and intermediary work in the adopting state
Amplified by: S02 Global funding seizure · A funding seizure removes the fiscal room for benefits, wage insurance and retraining.
S01 Trade bloc rupture · Bloc restrictions on models and chips leave some nations with the displacement through trade but not the adoption gain.
Amplifying trends: General-purpose AI capabilityAI inference efficiencyAI & knowledge-work automationFrontier model concentrationGraduate skill mismatchLifelong learning & digital reskillingEntrenched opportunity inequalityTax base erosion & coordination
Key trends: AI & knowledge-work automationCritical General-purpose AI capabilityCritical Social protection coverage gapsVery high Capital concentration in AIVery high AI safety & governanceVery high Digital & services trade growthVery high Frontier model concentrationVery high
From GINC Trends: the trends rated Very high or Critical for this scenario, of 250. All S08 trend scores.

13Rating rationale

Rating	Band or level	Why
Likelihood, two years	3Possible	METR measurements show the length of tasks that AI agents complete doubling about every seven months, and employment of young workers in exposed occupations has already fallen. A step that makes whole workflows viable within two years, with displacement on this scale following, would still require reliability, integration and liability barriers to fall together. Earlier general-purpose technologies took far longer to reach payrolls. Possible: 1 to 5 per cent.
Likelihood, ten years	4Likely	Over ten years, continued capability gains and falling inference costs make automation of whole workflows plausible. The uncertain part is the pace. Displacement of 10 per cent of employment within three years is much faster than past technological adjustment, and regulation, liability and organisational inertia slow adoption. One panel member rated this horizon Highly likely. The set holds it at Likely: 5 to 25 per cent.
Systemic impact	3Significant	Global output probably rises, so the rating rests on capability and not on output at risk. Band shifts are expected across human capital, public finance and cohesion in many nations at once. The loss falls on labour income, entry routes into skilled work and employment-linked revenue, and it is concentrated in high-income and services-exporting economies. Economies with small formal services sectors are less exposed. Significant, not Severe.
National impact	4Severe	For a highly exposed nation the highest dimensions are economic damage to households and social cohesion. One job in ten is removed or redesigned in three years, graduate entry halves and receipts from income tax and social contributions fall 10 per cent below plan while benefit demand rises. Concentrated displacement in the United States after China joined the WTO depressed local employment for more than a decade. Severe.
Confidence	low	The pace of AI capability and adoption is contested, the displacement share is an assumption and no historical event combines rising output with falling employment-linked revenue.

Source of scores: model-panel-v0.5. Confidence refers to the rating, not the scenario. Calibration sources are listed with the anchors above and on the scenario methodology page.

14Open questions

Contested assumptions for the panel to resolve.

Whether 10 per cent of employment in three years is the right Standard Severe quantity, given that no measured displacement from AI approaches it and the figure is an assumption of the panel.
How to rate systemic impact for an event that probably raises global output. The current rating rests on capability loss and not on output at risk.
Whether 'removed or radically redesigned' can be measured consistently across nations, or whether the test should use a narrower quantity such as net employment loss in exposed occupations.
Whether dependence on foreign model providers, dropped from the first-pass record, should return as a parameter or sit in a separate scenario.

15Commentary

No signed commentary in this build.

16Version and citation

Version: 0.5.0 · active · ai-labour-shock
Change log: 0.5.0 · 9 October 2026 · Rebuilt in the second pass from Rapid AI diffusion shock: narrowed to the labour-market face with a stated share of employment and period, onset moved from years to months, the market-concentration and provider-dependence variants removed; ten-year likelihood and systemic impact each lowered by one.
Full change log

Rating history

One row per edition. Each edition page is frozen at release; corrections are published as errata.

Edition	Rank	L2y	L10y	Systemic	National	Confidence	Edition view
2027	5	3Possible	4Likely	3Significant	4Severe	low	Open

Versions

Version	Date	Change
0.5.0	9 October 2026	Rebuilt in the second pass from Rapid AI diffusion shock: narrowed to the labour-market face with a stated share of employment and period, onset moved from years to months, the market-concentration and provider-dependence variants removed; ten-year likelihood and systemic impact each lowered by one.

Citation and downloads

Cite as

GINC (2027). Scenario S08 An AI deployment step causes abrupt labour displacement, Scenario Library v0.5. scenarios.ginc.org/scenarios/ai-labour-shock

Content and data are published under CC BY 4.0.

Download JSON Download CSV