Vulnerabilidad de Empleos a la Inteligencia Artificial en España: Dataset, Metodología y Dashboard Interactivo (v30 / v14)

AI Exposure of Jobs in Spain — Complete Dataset, Methodology & Interactive Dashboard (V30/v14) Overview Complete dataset, methodology, interactive dashboard, and adversarial reproducibility pack for analysing the vulnerability of 502 Spanish occupations to artificial intelligence. Built on the CNO-11 occupational taxonomy (SEPE expansion), EPA Q4 2025 microdata, Census 2021 structural weights, SEPE 2024 contract data, and administrative headcount registers for statutory public-service roles. The interactive dashboard is available at:  empleo-ai.anlakstudio.com Each of the 502 occupations receives an AI vulnerability score (0–10), employment estimate, salary estimate, EU AI Act regulatory classification, impact typology, four-dimensional sub-component scores (D/C/F/R), and a 3–4 sentence Spanish-language justification. Scores are calibrated for Spain using five structural modifiers (DESI digitalization, services sector weight, employment protection, EU AI Act, AESIA supervision). What's new in v30 (vs v29a) Methodology v29a → v30 (39 → 46+ pages, 56 → 57 technical notes, 6 → 7 appendices): New Appendix G — Full adversarial audit documentation. Documents the complete 3-wave red-team process: Wave 1 (structural integrity, v8, 7 simultaneous evaluators: Grok ×2, Perplexity ×2, Manus ×2, Gemini), Wave 2 (dimensional validation, v18–v27, GPT 5.4 + Gemini 3 + Claude + Manus + GPT-4o), Wave 3 (forensic artifact audit, v28, GPT-4o adversarial recomputation). Each wave with objectives, model roles, severity criteria, hallazgos tables, and resolution status. Includes risk matrix, comparative positioning table (this analysis vs Karpathy/Frey & Osborne/ILO/OECD on adversarial rigour), lessons learned, and reproducibility pack manifest. Note [57]: Reference to Appendix G and Zenodo reproducibility pack for the complete 28-issue register. Scoreboard: 24 of 28 issues resolved (86%). Remaining 4: 2 structural limitations (salary clustering, dual scoring regime), 1 blocked by data access (MCVL), 1 minor process. New supplementary file: REPRODUCIBILIDAD_RED_TEAM_v29a.zip (2.1 MB, 40 files) — full adversarial reproducibility pack enabling independent reproduction of all three red-team waves. Dataset remains v14 (unchanged from v29a). Issue tracker remains at 28 issues, 24 resolved. Key figures Metric Value Note Occupations analysed 502 CNO-11 complete (SEPE taxonomy) Workers represented 22.73 million EPA Q4 2025 + Census 2021 + SEPE 2024 + admin override Weighted mean vulnerability 3.7 / 10 Calibrated Spain (5 factors + sub-components) High vulnerability (≥7) 12.1% 47 occupations, 2.75M workers [≥6.5: 15.2% · ≥7.5: 11.7%] Vulnerability-weighted wage index 252,818M EUR Empleo × Salario × (Score/10) — index, not prediction Salary range 12,985 – 79,282 EUR/year EES 2023 + educational premium + FR/PT proxies (128 unique values) Inter-model validation (r) 0.715 100 occupations, Gemini 2.5 Pro vs GPT-4o blind Sub-component validation (502 occ) 100% D/C/F/R GPT 5.4 vs Gemini 3; 177 rescaled + 325 informative Salary validation (MAPE) 4.96% 16 groups INE EES 2023, post-correction Employment validation (1-digit) ±0.00% EPA Q4 2025 exact (API Tempus tabla 65134) Employment validation (2-digit) ±0.02% max drift Hard constraint in 62 EPA subgroups Unique employment values 499 / 502 4-layer cascade (EPA→Census→SEPE) + admin override Employment confidence A+ 2 occupations Admin-verified (headcount registers) Employment confidence A 385 occupations Census + SEPE chain Employment confidence B 115 occupations SEPE-synthetic chain Registered unemployed (SEPE) 10.02M (Dec 2024) Cross-validation signal: r = 0.73 vs contracts Karpathy US comparison 342 occ BLS vs 502 CNO-11 Appendix F: 16 dimensions, 9 notes [48–56] EU AI Act alto riesgo 51 occupations 3.30M workers (Annex III use-case mapping) Adversarial issues 28 identified, 24 resolved 3 waves, 7+ AI models, public issue tracker (Appendix G) Technical notes 57 Grouped by topic in Appendix A Dataset: spain_502_v14_subcomp_complete.json Records: 502 Format: JSON array Size: 807.8 KB MD5: 4439d959877e592668f8026c83c73a53 Fields per record: 34 (standard) / 41 (admin-override records, 5 occupations) Fields (34 standard) Field Type Description cno string 4-digit CNO-11 code (zero-padded) nombre string Official occupation name (Spanish) sector string 12 economic sectors empleo integer Employment estimate (4-layer cascade: EPA 1-dig → EPA 2-dig → Census 3-dig → SEPE 4-dig, or admin override) salario_medio_eur float Mean gross annual salary EUR (128 unique values; MAPE 4.96% vs INE EES 2023) vulnerabilidad_ia_score float AI vulnerability score (0–10, Spain-calibrated with 5 factors) eu_ai_act string Regulatory classification: "Alto riesgo," "Riesgo limitado," "Riesgo mínimo" tipo_impacto string Impact typology: "Sustitución" (29), "Híbrido" (184), "Aumentación" (289) justificacion string 3–4 sentence Spanish explanation of automation vectors and protective factors census_2021_employed float Census 2021 employment for intra-group weighting employment_method string Estimation method: epa1d_census3d_sepe4d, epa2d_hard, or admin_override| employment_confidence string Confidence tier: A+ (admin-verified), A (Census+SEPE), B (SEPE-synthetic) empleo_v10 integer Employment from v10 (audit trail) empleo_delta_v11 integer Employment change from v11 cascade update epa_2digit_empleo integer EPA 2-digit subgroup employment (hard constraint reference) score_v9 float/null Original v9 score before sub-component re-scoring rescore_method string sub-component-4D-rescaled (177 occ) or sub-component-4D-informative (325 occ) rescore_D float Sub-component: Digitalización (0–10) rescore_C float Sub-component: Cognitivo (0–10) rescore_F float Sub-component: Físico (0–10, higher = more physical barrier) rescore_R float Sub-component: Regulatorio (0–10, higher = more regulatory barrier) rescore_formula string Aggregation formula: (DC/10)(1-F/20)*(1-R/20) rescore_n_models integer Number of models used for sub-component scoring rescore_raw_avg float Raw average from formula before rescaling rescore_range string/null Score range across models (where available) rescore_cluster string/null Original cluster assignment (5.0 or 7.0, if applicable) flag_divergencia_gt2 boolean True if holistic–formula divergence exceeds 2.0 points (158 occupations) sepe_contracts_2024 integer SEPE registered contracts 2024 sepe_contracts_hombres integer Male contracts 2024 sepe_contracts_mujeres integer Female contracts 2024 sepe_parados_dic2024 integer Registered unemployed December 2024 sepe_parados_jun2024 integer Registered unemployed June 2024 sepe_parados_hombres_dic2024 integer Male unemployed December 2024 sepe_parados_mujeres_dic2024 integer Female unemployed December 2024 Additional fields (7, admin-override records only — CNO 5910, 5921, 5922, 5923, 5991) Field Type Description empleo_v13 integer Pre-override employment from v13 (audit trail) admin_source string Administrative data source name admin_source_ref string Specific table/page reference admin_source_url string URL to source publication admin_reference_date string Reference date of administrative data admin_confidence string Source reliability assessment (High/Medium) admin_notes string Context and cross-references Methodology: metodologia v30.pdf Pages: 46+ Technical notes: 57, grouped by topic (occupational inventory, employment, salary, AI scoring, EU AI Act, visualization, limitations, validation, calibration & adoption, Karpathy comparison, adversarial audit) Appendices: A (57 technical notes), B (inter-model validation with Bland-Altman charts), C (salary cross-validation, 16 INE EES 2023 groups), D (US comparison — methodology), E (quintile analysis), F (Karpathy vs de Nicolás: 16 dimensions across 9 notes [48–56]), G (Adversarial audit: 3 waves, 28 issues, 7+ models, risk matrix, scoreboard, reproducibility pack manifest — NEW in v30) Adversarial reproducibility pack: REPRODUCIBILIDAD_RED_TEAM_v29a.zip NEW in v30. 2.1 MB, 40 files. Enables full independent reproduction of all three red-team waves. Phase 1 — Structural (4 files): 5-phase adversarial protocol (300 lines), original v1 prompt (PDF), final consolidated prompt, roles_agentes_red_team.json with exact system prompts per model/role. Phase 2 — Multimodal (17 files): 10 raw scoring JSONs (4 batches cluster 7.0 + 6 batches cluster 5.0), 2 scoring prompt PDFs (GPT 5.4 + Gemini 3), processing script, changelog, 3 rescoring narratives, comparativa_inter_modelo.json with agreement metrics (r = 0.953 rescaled, r = 0.802 informative). Phase 3 — Forensic Final (7 files): Complete issue tracker (28 issues, 24 resolved), dataset v14 (502 occ, 22.73M workers), 2 consolidated 3rd-wave attack reports, v13_to_v14_patch_log.json (admin override for 5 security occupations), plausibility_gate.py (663 lines, stdlib only). Also contains: README.md and METADATA_ZENODO.json. Prior reproducibility package: prompts_and_scripts_v23.zip Previously uploaded (DOI: 10.5281/zenodo.19165098). Contains 22 files: 2 scoring prompt PDFs, 10 raw D/C/F/R JSON files (4 batch 7.0 + 6 batch 5.0), 3 batch instructions, processing script, changelog, build script, adversarial protocol, issue tracker, inter-model chart, and README. Interactive dashboard Production URL: empleo-ai.anlakstudio.com  Standalone file: empleo-ia-standalone_v30.html (self-contained React application) Features: Treemap (sector-level and occupation-level), scatter plot (salary × vulnerability), sortable table, detail panel with D/C/F/R sub-components, EU AI Act classification, employment confidence tier, impact typology, and justification text v29a UI features: 3-level methodology disclosure system (caveat pills → expandable limitations → full methodology section), inline wage-index disclaimer, separated EU AI Act badge with explainer, visible employment confidence tier Validation stack Employment (1-digit): EPA Q4 2025 totals from API Tempus (tabla 65134). Deviation: ±0.00% (max 47 persons over 22.46M base). Exact match at gran grupo level. Employment (2-digit): Hard constraint against 62 EPA 2-digit subgroups. Maximum drift: ±0.02% (was ±540% in v5 with Census-only weights). 4-layer cascade: EPA 1-digit hard → EPA 2-digit hard → Census 2021 3-digit proportional → SEPE 2024 4-digit contracts. Result: 499 unique employment values across 502 occupations. Employment (admin override): 5 occupations in CNO group 59 (security/penitentiary) patched with administrative headcount registers from Anuario MIR, Boletín AAPP, Idescat, INAP, and SGIP. +268,923 workers (+1.20% total). Each carries full audit trail with source, reference, URL, date, and confidence assessment. AI vulnerability scores (holistic inter-model): 100 stratified occupations blind-rescored by GPT-4o. Results: Pearson r = 0.715, ICC(2,1) = 0.701, weighted κ = 0.667. Bland-Altman: bias +0.28, 95% limits −2.77 to +3.33, no proportional bias. 84% agreement within ±2.0 points. AI vulnerability scores (sub-component D/C/F/R): 502/502 occupations scored on four dimensions (Digitalización, Cognitivo, Físico, Regulatorio) by GPT 5.4 and Gemini 3. 177 rescaled (r = 0.953 vs holistic), 325 informative (r = 0.802). Formula: (D×C/10) × (1−F/20) × (1−R/20). 158 occupations flagged with divergence > 2.0 points; 97% in direction formula < holistic. Salary (16 INE groups): Employment-weighted mean salary per EES 2023 group (tabla 28186) vs dataset estimates. Post-correction MAPE: 4.96%. Three groups corrected: Group I (protection/security) −36.4% → −4.4%, Group M (fixed machinery) −12.0% → −1.4%, Group H (health/care) +9.7% → +5.2%. Unemployment cross-validation: SEPE registered unemployed (parados) at 4-digit CNO, December and June 2024, with gender breakdowns. Contracts-parados correlation: r = 0.73. Adversarial multi-model review (3 waves — documented in Appendix G): Wave 1 (v8): 7 independent AI models simultaneously (Grok ×2, Perplexity ×2, Manus ×2, Gemini). 17 issues identified (5 critical, 9 significant, 3 minor). Wave 2 (v18–v27): GPT 5.4 + Gemini 3 as dimensional co-evaluators, Claude as synthesiser, Manus as source verifier, GPT-4o as holistic validator. Inter-model agreement improved from r = 0.715 to r = 0.953. Validation coverage: 20% → 100%. Wave 3 (v28): GPT-4o adversarial recomputation of public artifact bundle. 11 new issues (#18–#28), 2 critical (both FIXED). Combined: 28 issues identified, 24 resolved (86%). Red team verdict: "the methodology is defensible enough to publish." Sensitivity analysis Scenario Weighted mean High vuln. (≥7) Wage index Current calibration (base) 3.7 / 10 12.1% 252,818M EUR All factors −20% ~4.4 / 10 Non-linear est. ~303,400M EUR All factors +20% ~2.9 / 10 Non-linear est. ~202,300M EUR Threshold sensitivity (±0.5): high vulnerability ranges from 11.7% (≥7.5) to 15.2% (≥6.5) — a 3.5pp band, narrowed from 13.2pp pre-rescoring. EU AI Act classification (Regulation 2024/1689) Risk level Occupations Workers Examples Alto riesgo (Annex III) 51 3,297,718 Profesores, médicos, RRHH, policía, abogados, jueces Riesgo limitado 4 235,700 Teleoperadores, periodistas, traductores Riesgo mínimo 447 19,198,805 Camareros, albañiles, limpieza, operarios Note: The EU AI Act classifies AI systems by use-case (Annex III), not occupations. This analysis maps likely AI use-cases to the occupational contexts where they would deploy. Article 6(3) exemptions not captured; 51 is a conservative upper bound. Adoption context Scores measure theoretical AI capability, not current adoption. Three independent sources anchor the gap: INE TIC Q1 2025 (21.1% of firms use AI), Banco de España EBAE 2025 (~20%), Anthropic Economic Index (33% task-level in most-exposed occupations). Effective economy-wide task adoption: ~7%. The 5 calibration factors capture ~11–12% (structural friction); the remaining ~70pp reflects organisational barriers. Reader guidance: apply a 70–80% discount for current impact, varying by sector and firm size. Comparative positioning Dimension This analysis (v30) Karpathy 'Jobs' (US) Frey & Osborne (2013) OECD AI Exposure ILO GenAI Index Taxonomy CNO-11 (502, Spain) ONET/SOC (~800, US) SOC/O*NET (702, US) ~400 ISCO (cross-country) ISCO (cross-country) Scoring LLM + 5 calibration + 4D sub-components LLM single-pass (Gemini Flash) Expert panel (1 round) Expert + ONET tasks GPT-4 task scoring Inter-model validation r=0.715 holistic, r=0.953 sub-comp, Bland-Altman None published None published Expert panel None published Sub-component coverage 502/502 (100%) D/C/F/R None None None None Regulatory mapping EU AI Act (3 risk levels, 51 high-risk) None None None None Salary cross-reference Yes (128 values, MAPE 4.96%) Yes (BLS direct) No No No Unemployment data Yes (SEPE parados + gender, r=0.73 vs contracts) No No No No Adoption anchoring INE TIC 21.1% + BdE + Anthropic None None None None Adversarial audit 3 waves, 7+ models, 28 issues, public tracker None published None published None published None published Karpathy comparison Appendix F: 16 dimensions, 9 notes — — — — Documentation 46+ pp PDF, 57 technical notes, 7 appendices Code README Journal article Working paper Working paper US comparative analysis Parameter US (Karpathy) Spain (v30) Primary cause Mean vulnerability ~4.6 3.7 Physical services weight + 5-factor calibration + sub-components % high vulnerability (≥7) ~27% 12.1% [11.7%–15.2%] Smaller knowledge economy + sub-component rescoring Regulatory classification Not included 3 EU AI Act levels (51 high-risk) No US federal AI framework Labour friction factor Not applied 1–5% by sector OECD 3rd strictest employment protection Salary granularity ~800 direct values 128 adjusted values INE publishes at 2-digit level Employment granularity Direct per occupation 4-layer cascade: 499 unique values EPA anonymises CNO at 1–2 digit Sub-component rescoring Not included 502 occ × 4 sub-comp × 2 models v22–v30 Admin override Not applicable 5 occupations, +268,923 workers Statutory public-service headcount registers Empirical adoption anchor Not included INE TIC 2025: 21.1% Harmonised Eurostat survey Declared limitations Theoretical capacity, not displacement prediction. Scores measure what AI could do, not what it will do. The Anthropic Economic Index (Feb 2026) documents a gap: ~94% theoretical exposure vs ~33% observed adoption in computer/mathematical occupations. Employment at 4 digits is estimated. EPA publishes CNO at 1-digit only. 4-digit figures are proportional estimates via a 4-layer cascade. 385 occupations at Confidence A (Census+SEPE-validated), 115 at Confidence B (SEPE-synthetic), 2 at Confidence A+ (admin-verified). Scores are estimates (±0.5 intra-model reproducibility). Generated by Gemini 2.5 Pro at temperature 0.2, validated inter-model (r = 0.715, κw = 0.667) and via D/C/F/R sub-components (502/502). Bland-Altman confirms no proportional bias. Scores reflect March 2026 AI capabilities. Salaries are adjusted estimates. EES 2023 (reference year 2022) at 2-digit CNO, with educational premia and FR/PT intra-group proxies. 128 unique values. ~3.3M self-employed excluded by EES design. No temporal deflator applied (~+10% CPI 2022–2025). Salary clustering: 128 unique values; 445/502 occupations share a salary with at least one other. Root cause: INE publishes at 2-digit CNO only. Calibration factors are expert-informed. Five Spain-specific modifiers produce ~11–12% net score reduction. Interaction effects quantified (max 0.06 pts, negligible). Not empirically back-tested against observed adoption rates. Adoption gap documented with INE TIC 2025 (21.1%), Banco de España (~20%), and Anthropic (33% task-level). France/Portugal salary proxies assume structural similarity. MCVL is an identified but unused validation source. Sub-component regime is dual: 177 occupations have rescaled scores determined by D/C/F/R; 325 have informative sub-components that do not modify the holistic score. Field rescore_method distinguishes them. Static snapshot. March 2026. No job creation modelling, regional variation, or part-time/full-time distinction. The wage-vulnerability index (252,818M EUR) is NOT a prediction of wage losses or savings. It is a weighted concentration index. Do not cite as "X billion at risk." Data sources INE. Encuesta de Población Activa, Q4 2025. Microdatos (EPA_2025T4.tab, 107,497 registros). Publicados 27/01/2026. INE. Clasificación Nacional de Ocupaciones CNO-11. 449 grupos primarios oficiales. SEPE: 502 códigos (expansión). INE. Encuesta de Estructura Salarial 2023. Tabla 28186: salario bruto medio anual por subgrupo CNO. INE. Censo de Población y Viviendas 2021. Estructura ocupacional (3 dígitos CNO-11). INE. API Tempus, Tabla 65134: Ocupados por sexo y ocupación. Datos Q4 2025. SEPE. Estadísticas de contratos registrados 2024 y demandantes de empleo (diciembre 2024, junio 2024). Ministerio del Interior. Anuario Estadístico 2023 (Guardia Civil headcount). Ministerio de Hacienda. Boletín Estadístico del Personal al servicio de las AAPP, Enero 2024 (Policía Nacional headcount). Idescat, Parlamento Vasco, Gobierno de Navarra, Gobierno de Canarias (policías autonómicos). INAP. Revista INAP 2024 (policías locales estimate). SGIP / Academia de Prisiones (vigilantes de prisiones headcount). Comisión Europea. Digital Economy and Society Index (DESI) 2023. Perfil de España. Reglamento (UE) 2024/1689 del Parlamento Europeo y del Consejo (EU AI Act). Anexo III, Arts. 5 y 6. Real Decreto 729/2023. Estatuto de la AESIA. BOE de 2 de septiembre de 2023. INSEE (Francia). Données salariales par profession détaillée. Proxy salarial intra-grupo. INE-PT (Portugal). Quadros de Pessoal. Proxy salarial intra-grupo. INE. Encuesta sobre el uso de TIC y comercio electrónico en las empresas, Q1 2025 (21.1% use AI). Banco de España. EBAE (~20% firms use AI, 2025). OCDE. AI and Work. Employment Outlook 2024. Anthropic. The Anthropic Economic Index. Febrero 2026. Eloundou, T., Manning, S., Mishkin, P. y Rock, D. (2023). 'GPTs are GPTs.' OpenAI/UPenn. arXiv:2303.10130v5. Brynjolfsson, E., Mitchell, T. y Rock, D. (2018). 'What Can Machines Learn…' AEA Papers and Proceedings, 108:43-47. Frey, C. B. y Osborne, M. A. (2017). 'The Future of Employment.' Technological Forecasting and Social Change, 114:254-280. Acemoglu, D. (2024). 'The Simple Macroeconomics of AI.' Economic Policy. Felten, E., Raj, M. y Seamans, R. (2021). 'Occupational, industry, and geographic exposure to AI.' Strategic Management Journal, 42(12):2195-2217. Nedelkoska, L. y Quintini, G. (2018). 'Automation, skills use and training.' OECD Social, Employment and Migration Working Papers, No. 202. Karpathy, A. (2025–2026). 'Jobs.' BLS / O*NET analysis of US occupational AI exposure. Version history (this record) Version Date Dataset Key changes v17 2026-03-18 v5 Initial Zenodo publication v23 2026-03-20 v12 Reproducibility ZIP (22 files), prompts published v26 2026-03-21 v13 100% D/C/F/R validation, 56 technical notes v26B 2026-03-21 v13 Reproducibility ZIP expanded (42 files), full raw D/C/F/R outputs v27 2026-03-22 v13 Appendix F (Karpathy comparison), notes [48–56] v28 2026-03-23 v14 3rd red team mapped (11 issues), admin override for 5 security occupations v29 2026-03-23 v14 Appendix F terminology alignment (exposición → vulnerabilidad) v29a 2026-03-23 v14 UI deployment verified (4 issues fixed), 3-level disclosure system, 24/28 resolved v30 2026-03-23 v14 Appendix G: Full adversarial audit (3 waves, 28 issues, 7+ models). Reproducibility pack (40 files). Note [57]. 57 technical notes. 7 appendices Files File Description spain_502_v14_subcomp_complete.json Complete dataset (502 occupations, 34/41 fields, v14) metodologia v30.pdf Full methodology (46+ pages, 57 technical notes, 7 appendices incl. Appendix G) REPRODUCIBILIDAD_RED_TEAM_v29a.zip NEW — Adversarial reproducibility pack (40 files, 2.1 MB: prompts, raw scoring JSONs, issue tracker, dataset v14, plausibility gate, admin override patch log) prompts_and_scripts_v23.zip Reproducibility package (22 files: raw D/C/F/R scores, scoring prompts, batch instructions, processing scripts, adversarial protocol, issue tracker, README) empleo-ia-standalone_v30.html Interactive standalone dashboard Keywords artificial intelligence, labour market, employment, Spain, AI vulnerability, occupational risk, EU AI Act, CNO-11, EPA, automation, interactive dashboard, inter-model validation, Bland-Altman, treemap, wage-exposure index, AESIA, sub-component scoring, D/C/F/R, administrative data, police employment, SEPE, Census 2021, Karpathy comparison, unemployment, gender, adoption gap, reproducibility, adversarial audit, red team, radical transparency Zenodo Metadata Fields Resource type: Dataset + Interactive Visualisation + Methodology Document + Adversarial Reproducibility Pack Creators: Álvaro de Nicolás (ORCID: 0009-0004-8234-9720), Miguel Sureda (Affiliation: anlak) License: Creative Commons Attribution 4.0 International (CC BY 4.0) Language: spa (dataset, justifications, UI); eng (descriptions, methodology notes bilingual) Related identifiers: IsSupplementedBy: empleo-ai.anlakstudio.com (interactive dashboard, primary URL) IsSupplementedBy: zenodo.org (reproducibility package v23) IsSupplementedBy: github.com IsNewVersionOf: zenodo.org (previous version v29a/v14) References: Brynjolfsson, E., Mitchell, T. & Rock, D. (2018). "What Can Machines Learn, and What Does It Mean for Occupations and the Economy?" AEA Papers and Proceedings, 108:43-47. References: Eloundou, T., Manning, S., Mishkin, P. & Rock, D. (2023). "GPTs are GPTs." OpenAI/UPenn. arXiv:2303.10130v5. References: Frey, C. B. & Osborne, M. A. (2017). "The Future of Employment." Technological Forecasting and Social Change, 114:254-280. References: Regulation (EU) 2024/1689 (EU AI Act). Annex III, Arts. 5 and 6. References: Anthropic. The Anthropic Economic Index. February 2026. References: INE. Encuesta sobre el uso de TIC y del comercio electrónico en las empresas. Q1 2025. References: Banco de España. EBAE 2025. "La adopción de la inteligencia artificial en las empresas españolas." References: Nedelkoska, L. & Quintini, G. (2018). "Automation, skills use and training." OECD Social, Employment and Migration Working Papers, No. 202. References: Acemoglu, D. (2024). "The Simple Macroeconomics of AI." Economic Policy. References: Karpathy, A. (2025–2026). "Jobs." BLS / O*NET analysis of US occupational AI exposure. References: Felten, E., Raj, M. & Seamans, R. (2021). "Occupational, industry, and geographic exposure to AI." Strategic Management Journal, 42(12):2195-2217. Subjects: EuroSciVoc: Artificial intelligence (euroscivoc:297) EuroSciVoc: Employment (euroscivoc:1313) GEMET: Labour market (gemet:concept/4581)