Sources & Provenance

External citations, license, and contact for the synthesis. Per-field provenance is preserved in the underlying data file, in the _provenance block per country.

Version 1.1 · Last updated 2026-05-19

Based on 74 primary sources including Autor 2024 (QJE), Cedefop 2025, Draghi 2024, the European Climate Risk Assessment, Eurostat EUROPOP2023, the Rhodium / MIT Clean Investment Monitor, and OECD Pension Markets in Focus 2025.

1. How we built this The data-build pattern in two sentences.

Each of the 36 markets was scored across the five lenses, then placed into one of three corridors based on its Lens 1 ratio of displacement velocity to absorption capacity. Corridor placements were then stress-tested against the seven routine scenarios (with the eighth carried as a conditional), and the resulting trajectories sorted countries into the four fragility classes.

2. External Sources Cited Primary literature and institutional datasets referenced inline in the deliverable.

Tier 1Acemoglu, D., & Restrepo, P. (2019). Automation and New Tasks: How Technology Displaces and Reinstates Labor. Journal of Economic Perspectives.
Companion theoretical framework to Autor 2024 reinstatement-effect weakening; underpins the S1 New Jobs Replace Old and S6 New Jobs Fail to Appear scenario boundary. Source: link
Tier 1Acemoglu, D., & Restrepo, P. (2020). Robots and Jobs: Evidence from US Labor Markets. Journal of Political Economy.
Local-labour-market displacement evidence anchoring the C3 Displacement Without Absorption corridor. Sister evidence to El-Sahli & Upward 2017 for permanent earnings deficits. Source: link
Tier 1Autor, D., Chin, C., Salomons, A., & Seegmiller, B. (2024). New Frontiers: The Origins and Content of New Work, 1940–2018.
QJE 2024. Anchors the reinstatement-effect-weakening claim behind the corrected corridor edges; the primary citation for the S6 New Jobs Fail to Appear scenario.
Tier 1Bertheau, A., et al. (2022). The Unequal Consequences of Job Loss across Countries. IZA DP 15033.
Harmonised admin-data 5-year re-employment rates for AT/DK/FR/IT/PT/ES/SE. Direct input to the per-country A→C transition rates underpinning the Lens 1 absorption denominator. Source: link
Tier 1Bistline, J., et al. (2024). Power Sector Implications of the Inflation Reduction Act. NBER Working Paper 32168 / Brookings.
Academic anchor for IRA labour-outcome assessment; complements the Rhodium tracker and supports the “three years post-implementation confirms the mechanism class” claim in methodology SM 4. Source: link
Tier 1Brynjolfsson, E., Chandar, B., & Chen, R. (2025). Canaries in the Coal Mine? Six Facts about AI Employment Effects.
Documents ~16% relative employment decline for workers aged 22-25 in AI-exposed occupations (Sept 2025 ADP payroll data, controlling for firm-time effects). Informs the S3-vs-S6 scenario boundary and the speed-gap derivation (AI displacement 1–3 years against reskilling-system response 5–9 years); strengthens the Lens 1 entry-level signal alongside Massenkoff & McCrory 2026. Source: link
Tier 1Brynjolfsson, E., Li, D., & Raymond, L. R. (2023). Generative AI at Work. NBER Working Paper 31161.
RCT in customer-support work showing 14% productivity gain from generative AI assistance, concentrated in less-experienced workers. Direct anchor for the S3 Jobs Transform scenario (within-occupation task augmentation rather than between-occupation displacement). Source: link
Tier 1Card, D., Kluve, J., & Weber, A. (2018). What Works? A Meta-Analysis of Recent ALMP Evaluations. Journal of the European Economic Association.
Meta-analytic evidence on Active Labour Market Programme effectiveness. Calibrates the absorption-capacity denominator in the Lens 1 ratio and anchors the S7 Bandwidth Fracture scenario when ALMP capacity is crowded out. Source: link
Tier 1Cedefop (2025). European Skills and Jobs Survey + Skills Forecast 2025.
Country-level employment projections; Zone-A vs Zone-C task-substitution split; baseline for the S2 Climate Adaptation Boom mass.
Tier 1Dauth, W., Findeisen, S., Südekum, J., & Wößner, N. (2021). German Robots: The Impact of Industrial Robots on Workers. Journal of the European Economic Association.
Country-specific evidence for the Continental sub-cluster mechanism (worker-protection × tech-substitution). Anchors the DE / AT fragility-class assignments and the within-cluster mobility-versus-protection trade-off. Source: link
Tier 1Draghi, M. (2024). The Future of European Competitiveness.
European Commission. Strategic anchor for European competitiveness diagnosis. Direct overlap with Lens 5 institutional adaptive-capacity scoring; the productivity-gap framing supports the strict-zero finding (Europe's capacity to absorb structural shocks has degraded versus the US over 20 years). Bridges to Part 7's Tier 1 framing. Source: link
Tier 1EEA: European Climate Risk Assessment (ECRA) 2024.
Compounding-crisis (Lens 4) component; eea_vuln readings for Class IV cascade signal.
Tier 1El-Sahli, Z., & Upward, R. (2017). Off the Waterfront: The Long-Run Impact of Technological Change on Dockworkers.
Anchors the C3 Displacement Without Absorption corridor: structural lifetime-earnings deficits among displaced workers.
Tier 1Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs: Labor Market Impact Potential of LLMs.
Theoretical AI exposure ceiling per occupation (E1+E2 scores). Third index in the Lens 1 triangulation alongside the Anthropic Economic Index and Microsoft Working with AI. Source: link
Tier 1EU MFF Mid-Term Review 2024.
€64.6 B reinforcement (Ukraine + migration + emergency + STEP); aggregate-level signal for the Lens 5 bandwidth-allocation proxy. Per-country disaggregation is in the known-limits list.
Tier 1EU Net-Zero Industry Act & Clean Industrial Deal (Feb 2025).
€100 B clean-manufacturing envelope; the wage-positive Zone-C premium that makes S2 the only routine path to C1 for AT, LU, TR.
Tier 1European Commission (2024). 2024 Ageing Report: Economic and Budgetary Projections for the EU Member States (2022–2070).
Working-age and dependency-ratio projections; cross-feeds Lens 2 demographic-buffer scoring and the Lens 5 fiscal-bandwidth context for defence + climate + migration commitments. Source: link
Tier 1Eurostat: Adult Education Survey (trng_aes_100).
Country-level non-formal adult-learning participation. Speed-axis indicator for the per-country reskilling-system radar feeding Lens 1 absorption and the per-country reskilling-capacity gap. Source: link
Tier 1Eurostat: Employment by ISCO 2-digit (lfsa_egai2d / lfsa_egais).
Backbone employment table for the Klinger 2-digit ISCO coordination-share weighting and for the high-exposure cohort identification feeding Lens 1. Source: link
Tier 1Eurostat: Employment rate of older workers, 55–64.
Country-level older-worker employment baseline. Direct input to the Lens 2 retirement-offset calculation and to the demographic-buffer ceiling for each market. Source: link
Tier 1Eurostat: Enterprise AI adoption (isoc_eb_ai).
Per-country share of enterprises (10+ employees) using at least one AI technology, 2021–2025. Direct input to per-country Lens 1 displacement-velocity calibration. Source: link
Tier 1Eurostat: EUROPOP2023 + lfsa_etpgan tenure-under-1yr proxy.
Demographic projections; baseline-economic-churn estimate for the reskilling-capacity calculation.
Tier 1Eurostat: ICT specialists as share of employment (isoc_sks_itspt).
Country-level ICT-specialist share, 2015–2024. Anchors the ICT-heavy weight (0.157) in the high-coordination cluster split that surfaces the teaching-heavy versus tech-heavy archetype divergence. Source: link
Tier 1Feigenbaum, J., & Gross, D. (2024). Automation and the Fate of Young Workers: Evidence from Telephone Operation. Quarterly Journal of Economics.
Historical evidence on entry-level displacement following automation; calibration anchor for the S6 New Jobs Fail to Appear scenario and the entry-level signal flagged in the speed-gap derivation. Source: link
Tier 1Handa, K., Tamkin, A., et al. (2025). The Anthropic Economic Index: observed Claude usage by occupation.
SOC-mapped task-coverage series. Observed AI usage anchor for Lens 1 displacement velocity; the bottom-heavy distribution (4% of occupations show usage on ≥75% of tasks) is the empirical floor for the S3 Jobs Transform scenario. Source: link
Tier 1IISS Military Balance 2025.
Defence-spending share of GDP; Lens 5 polycrisis-drag input.
Tier 1ILO (2025). Generative AI and Jobs: A Refined Global Index of Occupational Exposure.
Third independent AI-exposure index alongside Anthropic and Microsoft. Cross-check on the Lens 1 displacement-velocity ceiling for cross-country deployment-pattern divergence. Source: link
Tier 1International Monetary Fund (2026). World Economic Outlook Database.
Country-level GDP and fiscal indicators feeding the Lens 5 defence-spending share and bandwidth-allocation proxy. Source: link
Tier 1Jacobson, L., LaLonde, R., & Sullivan, D. (1993). Earnings Losses of Displaced Workers. American Economic Review.
Foundational earnings-loss estimates for displaced workers. Sister anchor to El-Sahli & Upward 2017 for the C3 corridor lifetime-deficit signal. Source: link
Tier 1Massenkoff, M., & McCrory, P. (2026). Labor Market Impacts of AI: A New Measure and Early Evidence.
Anthropic Research. Earliest large-sample evidence on AI hiring effects per occupation. Feeds Lens 1 displacement-velocity calibration and the entry-level signal flagged in the L5 Speed-Gap derivation. Source: link
Tier 1Munich Re NatCat 2025 (Europe).
Lens 4 climate-shock count; Tier 1 reinsurance-grade NatCat dataset.
Tier 1NATO Hague Summit Declaration (June 2025) · ReArm Europe / Readiness 2030.
Defence-spending uplift commitments; Lens 5 institutional-bandwidth tax.
Tier 1OECD (2025). Pension Markets in Focus 2025.
Per-country pension-asset operationalisation evidence for the Wealth-Fund-Rich regime hypothesis (declined 2026-05-08 on threshold-inconsistency grounds); cited in methodology SM 4 v5 second paragraph. Source: link
Tier 1OECD Economic Surveys: European Union and Euro Area 2025.
Country-level institutional and fiscal capacity assessment. Anchors the Lens 5 cross-country comparison and the post-growth-empirical regime classification (10 markets). Source: link
Tier 1OECD Employment Protection Legislation (EPL) Database.
Country-level EPL strictness scores (regular contracts, 0–6 scale). Direct input to Lens 5 jurisdictional-buffering scoring; underpins the worker-protection axis of the eight-country squeeze cluster. Source: link
Tier 1OECD: Old-age dependency ratio.
Country-level old-age dependency ratio. Direct input to Lens 2 demographic-buffer scoring and to the retirement-offset calculation that bounds the maximum-buffer reading at ~26%. Source: link
Tier 1OECD: Social Expenditure Database (SOCX), ALMP training-category spend % GDP.
Replaces the withdrawn Eurostat empl_lmp_expsumm series. Per-country ALMP spending intensity feeds the Lens 1 absorption-capacity scoring and the S7 Bandwidth Fracture scenario. Source: link
Tier 1Rhodium Group / MIT: Clean Investment Monitor (IRA tracker).
Realised US labour data on state-coordinated industrial buildout; cross-jurisdictional confirmation of the Climate Adaptation Boom (S2) mechanism class. The Q1 2025 fragility footnote ($8 B / 27,000 jobs cancelled) anchors the “mechanism class confirmed, fragility documented” framing in methodology SM 4. Source: link
Tier 1SIPRI Trends in World Military Expenditure 2025.
Ukraine reference panel anchor (USD 84.1 bn 2024 = 40 % of GDP); Lens 5 upper-bound calibration.
Tier 1Tomlinson, K., Jaffe, S., et al. (2025). Working with AI: Measuring the Applicability of Generative AI to Occupations.
Microsoft Research. Capability-ceiling counterpart to the Anthropic observed-usage series; the two together produce the capability-vs-deployment split used for Lens 1 dual-baseline scoring. Source: link
Tier 1United Nations Population Division (2024). World Population Prospects 2024.
Global comparator for the European demographic trajectory; provides the EU and per-country baseline projections used as a check on EUROPOP2023 alignment. Source: link
Tier 1Atkinson, T., & Yamco, S. (2026). Young workers' employment drops in occupations with high AI exposure.
Federal Reserve Bank of Dallas, January 6, 2026. Documents the inflow-vs-outflow mechanism behind the entry-level signal: job-finding rate down more than 3 percentage points since the November 2023 peak; separation rates flat. Source of the developmental-displacement-undercount caveat in §8.7. Source: link
Tier 1Liu, J., & Webber, D. A. (2026). AI Adoption and Firms' Job-Posting Behavior.
FEDS Notes, Federal Reserve Board, March 27, 2026. Precisely-estimated null effects on industry- and firm-level job postings across 7.3 million observations and 1.05 million firms. Anchor for the unit-of-analysis disambiguation in §4.6: vacancies show no AI-driven decline. Source: link
Tier 1Audoly, R., Guerin, M., & Topa, G. (2026). Do Job Postings Show Early Labor-Market Effects of AI?
Liberty Street Economics, Federal Reserve Bank of New York, May 14, 2026. Lightcast-based replication of the vacancy null. Confirms that any relative posting decline in high-exposure occupations predates ChatGPT. Co-anchor for the §4.6 unit-of-analysis disambiguation. Source: link
Tier 1Aldasoro, I., Gambacorta, L., Pál, R., Revoltella, D., Weiss, C., & Wolski, M. (2026). AI adoption, productivity and employment: Evidence from European firms.
BIS Working Paper 1325 / CEPR Discussion Paper 21082 / EIB Working Paper 2026/02, January 23, 2026. First causal-design evidence at firm level across 12,000 EU and US firms: +4% labour productivity, no adverse firm-level employment effect. Anchor for the European firm-level nulls strand of the §4.6 disambiguation. Source: link
Tier 1Lebastard, L., & Sondermann, D. (2026). Artificial Intelligence: friend or foe for hiring in Europe today?
ECB Blog, March 4, 2026. Survey of 5,000 SAFE-surveyed euro-area firms; AI-intensive firms approximately 4% more likely to hire. Supplements Aldasoro and others as the SAFE-based European firm-level null. Source: link
Tier 1Lane, P. R. (2026). AI and the euro area economy.
ECB speech, ECB-SAFE-RCEA International Conference (3CMFI), Frankfurt, March 23, 2026. ECB-official synthesis: “currently little evidence of a substantial effect of AI on employment in the euro area.” Anchors the institutional reading that pairs with Lebastard & Sondermann and Aldasoro and others. Source: link
Tier 1Humlum, A., & Vestergaard, E. (2025). Large Language Models, Small Labor Market Effects (subsequently retitled “Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI”). NBER Working Paper w33777.
Danish administrative-data study, 25,000 workers across 7,000 workplaces and 11 AI-exposed occupations. Precise nulls on earnings and hours two years post-ChatGPT, with the additional observation that the Danish early-career employment decline does not correlate with which firms adopted AI. Key adoption-decoupling input to the §4.6 disambiguation. Source: link
Tier 1Jaumotte, F., Kim, J., Koll, D., Li, E. Z., Li, L., Melina, G., Song, A., & Mendes Tavares, M. (2026). Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age. IMF Staff Discussion Note SDN/2026/001.
January 9, 2026. Cross-country anchor across BRA, DNK, DEU, GBR, USA, ZAF. Finds 3.6% lower employment in regions with greater AI-skill demand for occupations highly exposed but with limited complementarity. Explicit “steppingstone” framing for young workers in the career ladder. Source: link
Tier 1OECD / Korea Labor Institute (2025). Artificial Intelligence and the Labour Market in Korea.
OECD Publishing, 2025. Cross-country OECD synthesis (Chapter 2): firms in eight OECD economies (finance and manufacturing, 2022) “often opted to slow hiring instead of implementing job cuts” when introducing AI. Korea-specific finding (Table 2.3): traditional-AI exposure associated with employment-growth declines concentrated among younger workers, low-to-medium-skilled workers, and women. Institutional endorsement of the §4.6 inflow mechanism plus non-Western corroboration. Source: link
Tier 1IAB Institut für Arbeitsmarkt- und Berufsforschung (2025). IAB-Kurzbericht 19/2025: Arbeitsmarktprognose 2025/2026.
Forecasts German employment to decline by approximately 20,000 persons in 2026; Erwerbspersonenpotenzial declines for the first time since the pandemic. Cyclical confounder for the German country row (now in countries.DE.cyclical_confounder_2026). Source: link
Tier 1NIESR (2026). Economic Outlook Winter 2026: Normality Under Strain.
National Institute of Economic and Social Research. Calculates that employer NIC increases, NLW upratings, and rights reforms together raised the real marginal cost of entry-level hiring in the UK by approximately 7% year-on-year. Policy confounder for the UK country row (now in countries.UK.policy_confounder_2026). Source: link
Tier 1Serôdio, P. (2026). AI and the UK labour market: the evidence so far.
British Progress, April 22, 2026. Three years post-ChatGPT, no detectable UK-wide AI employment effect at occupation level. Theoretical-vs-observed exposure gap consistent with Massenkoff & McCrory. Cross-reference for the UK policy-confounder framing. Source: link
Tier 1Kilincarslan, E., & Li, J. (2026). What impact is AI having on British firms and the jobs they offer?
LSE Business Review, March 19, 2026. UK unemployment 5.2% (three months to January 2026); junior and administrative roles bear the AI-adoption brunt according to surveyed employer expectations. Sector-anecdote support for the UK policy-confounder framing. Source: link
Tier 1Bailey, A. (2025). Bank of England Governor on AI and the entry-level career pipeline.
BBC Radio 4 Today programme interview, December 19, 2025; original broadcast no longer available on BBC Sounds. Quoted in Insurance Journal, eweek, and bmmagazine, 2025-12-19. “There is an issue with younger, inexperienced professionals finding it difficult to secure entry-level roles due to AI.” UK-institutional acknowledgement of the inflow framing. Source: link
Tier 1Engberg, E., Görg, H., Hellsten, M., Javed, F., Lodefalk, M., Längkvist, M., Monteiro, N., Kyvik Nordås, H., Pulito, G., Schroeder, S., & Tang, A. (2026). Who is afraid of AI? Who should be? Kiel Policy Brief No. 198.
Kiel Institute for the World Economy, January 2026. Three-country firm-level analysis (Denmark, Portugal, Sweden) using a new Dynamic AI Occupational Exposure (DAIOE) measure with 9 AI subdomains 2010-2023. Finds no systematic change in total firm employment, with a clear positive association between AI exposure and the skill ratio across all three countries. Subdomain heterogeneity: reading-comprehension, speech-recognition, and language-modelling exposure are net-positive for total employment. Strengthens the European firm-level nulls cluster in §4.6 with an explicit upskilling framing. Source: link
Tier 2Acemoglu, D., & Johnson, S. (2023). Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity.
Strategic context for technology distribution outcomes; informs the Lens 5 institutional-direction question that distinguishes routine from off-path scenarios. Source: link
Tier 2Allianz Research (2024). European labor markets: Migration matters.
Migration buffer analysis informing the Lens 2 retirement-offset ceiling. Source: link
Tier 2Autor, D., Dorn, D., & Hanson, G. (2013). The China Syndrome. American Economic Review.
Local-labour-market displacement-without-absorption analytical framework; informs the C3 corridor mechanism and the regional-reabsorption gap diagnosed in the C2 Partial Absorption corridor. Source: link
Tier 2Bruegel (2025). The Demographic Divide: Inequalities in Ageing Across the European Union.
Cross-country demographic-asymmetry framing. Informs the Lens 2 per-country buffer split and the Italy / Greece secular-stagnation-warning regime classification. Source: link
Tier 2Brynjolfsson, E., Rock, D., & Syverson, C. (2021). The Productivity J-Curve. American Economic Journal: Macroeconomics.
Capability-versus-deployment lag framing; informs the Lens 1 dual-baseline (capability ceiling vs observed usage) and the S3 Jobs Transform within-occupation reinstatement timing. Source: link
Tier 2Cooke, R. M. (1991). Experts in Uncertainty: structured-elicitation framework.
Methodological anchor for the 80 % CI bands per scenario × regime in the probability vectors.
Tier 2ETUC (2026). Press release on the EU Industrial Accelerator Act, 2026-03-04.
Labour-side framing of the Industrial Accelerator Act; convergent with IndustriAll Europe on works-council mediation and procurement-attached social conditionalities. Anchors the S2 mechanism-string additive (data.json line 7941). Source: link
Tier 2EU AI Act: Regulation (EU) 2024/1689.
Phased in force August 2025–August 2027. Lens 5 regulatory anchor; underpins the AI-governance hiring intent surfaced in S3 Jobs Transform and the high-risk classification threshold informing per-country compliance bandwidth. Source: link
Tier 2EURES / European Labour Authority (2024). Report on Labour Shortages and Surpluses in Europe 2024.
Sector-level shortage signals informing the Lens 2 labour-buffer scoring and the per-country Zone-A versus Zone-C task-substitution split feeding S2 Climate Adaptation Boom. Source: link
Tier 2European Parliament (2025). Displaced Ukrainians: Challenges and Outlook for Integration in the EU.
Migration-bandwidth context for the Lens 4 polycrisis-drag and Lens 5 jurisdictional-buffering inputs; informs the EU MFF Mid-Term Review €64.6 B reinforcement signal. Source: link
Tier 2Feigenbaum, J., & Gross, D. (2025). Organizational and Economic Obstacles to Automation: A Cautionary Tale from AT&T. Management Science.
Slow-deployment evidence within a single firm and technology cycle; informs the S4 Muddle Through and S7 Bandwidth Fracture scenario timing assumptions. Source: link
Tier 2Frey, C. B. (2019). The Technology Trap: Capital, Labor, and Power in the Age of Automation.
Long-run displacement-without-absorption analytical anchor; informs the S6 New Jobs Fail to Appear and S7 Bandwidth Fracture scenarios. Source: link
Tier 2Hall, P., & Soskice, D. (2001). Varieties of Capitalism. Oxford University Press.
Institutional-typology framework anchoring the fragility-class taxonomy and the Nordic / Continental / Liberal sub-cluster split inside the eight-country squeeze cluster. Source: link
Tier 2IndustriAll Europe (2026). Article 1450: Made in Europe 2.0: labour-side reception.
Sectoral-union framing complementary to ETUC; second source for the S2 additive convergence on works-council mediation and procurement-attached social conditionalities. Source: link
Tier 2IPCC AR6 Likelihood Scale.
Probability vocabulary (very likely, likely, about as likely as not, unlikely, very unlikely) used in per-country corridor-mass distribution sentences.
Tier 2Klinger coordination-share weighting (2-digit ISCO with ESCO occupation counts).
Methodology used to surface the teaching-heavy (weight 0.582) vs ICT-heavy (weight 0.157) archetype split inside the high-coordination cluster.
Tier 2OECD (2023). Pensions at a Glance 2023.
Cross-country pension-system comparison feeding the Lens 5 fiscal-bandwidth context and the post-growth-empirical regime classification (10 markets). Source: link
Tier 2Tooze, A. (Chartbook 130, 2022 · Chartbook 407, 2025).
Framing reference for the framing-vs-mechanism distinction the Lens 5 spec deliberately separates from.

3. Sister-layer source bases Where each upstream layer's primary evidence sits.

The synthesis pulls from each of the five sister layers. Each maintains its own primary-source registry; this section links out rather than duplicating.

Primary legal texts plus AI exposure indices: Anthropic Economic Index, Microsoft Working with AI, ESCO, Cedefop, O*NET, plus EU and DACH regulation.
26 primary sources covering European job-market signals: Eurostat, OECD, Ravio, LinkedIn, Stack Overflow.
52 sources across three tiers behind 20 case studies of technology disruption, spanning peer-reviewed econometrics through secondary journalism.
54 primary sources from Eurostat, national statistical offices, the European Commission, and academic research.
63 sources across three quality tiers, each annotated to a headline claim and cross-checked against the Derivation Appendix.

License

Data & Analysis: CC-BY 4.0. You may share, adapt, and build on the work, including for commercial purposes, with attribution.

Code: MIT.

Cite as: Maul, P. (2026). Part 6 · European AI Labour Market Synthesis. Nexalps. https://synthesis.nexalps.com/

4. Contact & Feedback

Found a methodological issue? A miscoded country? An obviously missing source? Reach out via LinkedIn.

LinkedIn · Philipp Maul Nexalps