AI-Exposed Occupations

Updated 10 January 2026

AI-exposed occupations are roles susceptible to automation or augmentation by AI systems, defined by advanced task analysis.
Methodologies like the Occupational AI Exposure Score and patent–task matching quantify exposure by assessing task overlaps with AI capabilities.
Empirical findings show that high exposure in knowledge-intensive roles correlates with changes in employment, wage structures, and required skills.

AI-exposed occupations are defined as occupational roles in which the task content is, in principle, susceptible to automation or augmentation by contemporary AI systems, particularly LLMs and multimodal generative models. Exposure is quantified via the potential or realized overlap between occupational tasks and AI capabilities, often operationalized through task-level measurement with respect to existing or anticipated system performance. While exposure does not guarantee substitution or displacement, it is a necessary precondition for any labor-market transformation due to AI. Multiple methodological frameworks—task-based simulation with cutting-edge LLMs, patent–task matching, labor-market outcome regressions, and market-adoption tracking—have converged on quantitatively robust but heterogeneous patterns of AI exposure across sectors, occupations, demographics, and geographies.

1. Methodologies for Quantifying Occupational AI Exposure

Task-based quantification dominates recent measurement regimes. A leading example is the Occupational AI Exposure Score (OAIES), constructed by prompting LLMs (e.g., ChatGPT-4o, Claude 3.5 Sonnet) to estimate the share of O*NET task content automatable at discrete "AI capability stages"—from pre-LLM ML/NLP, to early and multimodal LLMs, to advanced reasoning and agentic AI. OAIES for occupation $o$ at stage $s$ is defined as

$\mathrm{OAIES}_{o,s} = \frac{\sum_{t=1}^{T_o} r_{t,o}\,P_{t,o,s}}{\sum_{t=1}^{T_o} r_{t,o}},$

where $r_{t,o}$ is O*NET's relevance weight for task $t$ in occupation $o$ , and $P_{t,o,s}$ is the percent of task $t$ 's subtasks estimated automatable at stage $s$ (Dominski et al., 11 Jul 2025).

Complementary approaches include:

Patent–task matching: The AI Impact (AII) framework uses deep learning to measure semantic similarity between U.S. AI patent abstracts and O*NET tasks. The occupational AII score is the fraction of tasks with high similarity to AI patents (Septiandri et al., 2023).
Disruption/consolidation indices: Task-level exposure is further partitioned by the disruptive/consolidative character of AI innovations, leveraging patent citation network disruption scores to classify whether an occupation's tasks are targeted by status quo-reinforcing or frontier-shifting AIs (Kim et al., 15 Jul 2025).
Startup exposure: The AI Startup Exposure (AISE) index uses binary LLM assessment of which O*NET-described job tasks are actively targeted for automation by products from AI startups (as catalogued in Y Combinator archives), capturing contemporaneous application rather than technical feasibility (Fenoaltea et al., 2024).
Direct user-activity mapping: AI applicability scores can be calculated by analyzing anonymized user–LLM interaction logs to determine which work activities are performed or assisted by AI and with what effectiveness, then aggregated by occupational work-activity composition (Tomlinson et al., 10 Jul 2025).

Robustness is enhanced by model ensembles (Frank et al., 2023), theory-driven annotation of fundamental automatizability grounded in Moravec's Paradox (Schaal, 15 Oct 2025), and scenario-based simulation at scale (e.g., the Iceberg Index) (Chopra et al., 29 Oct 2025).

2. Empirical Patterns of High and Low AI Exposure

Task-based and empirical exposure rankings consistently show:

High exposure for knowledge-intensive, data-driven, and cognitively complex roles:
- Software Developers, Applications, Data Scientists, Financial Analysts, Management Analysts, Accountants and Auditors, Market Research Analysts, Writers and Authors, Paralegals and Legal Assistants, Customer Service Representatives (Dominski et al., 11 Jul 2025, Septiandri et al., 2023, Kim et al., 15 Jul 2025, Chopra et al., 29 Oct 2025).
- Interpreters and Translators, Editors, Historians, Postsecondary Teachers (especially English/language/history), Technical Writers, Credit Analysts (Felten et al., 2023, Tomlinson et al., 10 Jul 2025).
Low exposure for roles dominated by:
- Manual/physical activity (Locksmiths, Dining Attendants, Equipment Cleaners, Construction Laborers, Agricultural Equipment Operators).
- On-site variability or high sensorimotor skill loads (Riggers, Industrial Machinery Mechanics, Roustabouts).
- Regulatory/ethical or interpersonal requirements (Magistrate Judges, Nurse Anesthetists, Clergy, Healthcare Support) (Fenoaltea et al., 2024, Schaal, 15 Oct 2025).

This stratification is robust across diverse models (OAIES, AII, AISE, Iceberg, TEAI/ TRAI), with the strongest convergence observed for STEM, business/finance, and high-wage administrative roles, and the weakest for low-skill physical/labor and highly embodied caregiving, educational, or creative (in the sense of visual/performative arts) work (Dominski et al., 11 Jul 2025, Schaal, 15 Oct 2025).

Rank	High-Exposure Occupation	Exposure Metric	Low-Exposure Occupation	Exposure Metric
1	Software Developers, Applications	Iceberg 78.5%	Riggers	Theory-based 0.15
2	Data Scientists	Iceberg 74.3%	Dining Attendants/Bartender Helpers	OAIES Δ≈8.4
3	Accountants and Auditors	Iceberg 65.8%	Construction Laborers	AII, Theory-based

3. Labor Market Effects and Heterogeneous Impact

AI exposure is empirically associated with both negative and positive labor-market consequences, with strong heterogeneity by task content, demographic group, and temporal lag:

Negative employment outcomes: From late 2022 to early 2025, high-exposure occupations experienced significant declines in employment (–5.6 to –8.5 percentage points per 10-point exposure increase), higher unemployment rates, reduced main-job hours, and lower full-time shares. These effects are accentuated for occupations with:
- High nonroutine cognitive–analytical task content (e.g., business analysts, editors, news reporters).
- Workers ≤30 or ≥50 years old, men, and those with higher education levels, though college-educated experienced smaller employment losses but disproportionately greater drops in hours and full-time rates (Dominski et al., 11 Jul 2025).
Manual/physical and routine-manual occupations frequently exhibit resilience or slight employment gains as AI exposure increases, attributed partly to slow technical progress in embodied AI/robotics and complementarity with cognitive automation (Dominski et al., 11 Jul 2025, Kim et al., 15 Jul 2025).
Wage dynamics: Evidence indicates a positive relationship between AI exposure and wages, with more exposed roles typically commanding higher pay. However, wage premiums are compressing post-LLM, and increased exposure—if not accompanied by upskilling—may reverse historical skill-biased technological change (Schaal, 15 Oct 2025, Henseke et al., 30 Jul 2025).
Demand for new (complementary) work created by augmentation AI is observed in high- and medium-skill occupations, while automation AI suppresses new work and wage growth in low-skill roles (Marguerit, 24 Mar 2025).

4. Exposure Typologies: Automation vs. Augmentation, Disruption vs. Consolidation

AI "exposure" metrics distinguish several axes:

Automation Exposure: Potential for AI to substitute human labor in current tasks; highest in routine cognitive, administrative, and information processing roles (Marguerit, 24 Mar 2025).
Augmentation Exposure: Potential for AI to enhance or complement human labor—through workflow improvement, decision-support, or knowledge transfer; most salient in STEM, professional, and creative problem-solving domains.
Disruptive AI Innovations: Target unpredictable, mental, and analytical tasks, disproportionately impacting science/technology sectors in coastal regions and areas with labor shortages (Kim et al., 15 Jul 2025).
Consolidating AI Innovations: Reinforce automation of physical/routine/solo tasks, affecting manufacturing and construction in the Midwest and similar geographies (Kim et al., 15 Jul 2025).

Notably, occupation-level exposure is increasingly multidimensional—combining task automability, complementarity with AI, market viability, and societal regulation or acceptability (Fenoaltea et al., 2024, Schaal, 15 Oct 2025).

5. Sectoral, Educational, and Geographic Variation

Sectoral and demographic stratification is pronounced:

Sectors: Highest exposures cluster in professional, scientific, technical services (OAIES, AISE, Iceberg), IT, finance, legal, healthcare information management, and administrative support. Manufacturing and logistics remain more exposed to consolidating (robotic) automation (Septiandri et al., 2023, Chopra et al., 29 Oct 2025).
Education: High-exposure roles are concentrated among bachelor’s, master’s, and doctoral holders (percentile rank: BA ~60th, MA ~80th, PhD ~90th), though entry-level and less-educated workers face greater employment risk within their exposure group (Tomlinson et al., 10 Jul 2025, Dominski et al., 11 Jul 2025).
Geography: Hotspots for disruptive AI exposure are found in coastal states and major metro tech hubs; consolidating exposure is higher in Midwest and central regions dense in manufacturing, fabrication, and construction (Kim et al., 15 Jul 2025, Chopra et al., 29 Oct 2025).
Temporal evolution: Aggregate exposure has risen since 2017, but the main driver is compositional employment shifts toward more susceptible occupations, rather than drastic within-occupation task change (Henseke et al., 30 Jul 2025).

6. Implications for Workforce Policy and Adjustment

The empirical findings motivate several strategic responses:

Targeted reskilling and upskilling in knowledge-intensive, highly exposed roles, with urgent deployment of digital-literacy and cognitive-automation training both in educational and industrial policy (Chopra et al., 29 Oct 2025, Henseke et al., 30 Jul 2025).
Workforce planning needs to recognize that augmentation is the dominant pattern in many high-skill exposures, requiring a reframing of curricula and job design to integrate human–AI collaboration, quality control, and ethical oversight skills (Tomlinson et al., 10 Jul 2025).
Early monitoring and adaptive intervention are necessary, as labor-market effects are already observable in near real-time; both wages and job transitions demand integrated indices (OAIES, Iceberg, GAISI) for ongoing risk and opportunity assessment (Dominski et al., 11 Jul 2025, Chopra et al., 29 Oct 2025).
Policy granularity should increase: dynamic, composite, and region- or sector-specific indices outperform single-model rankings in predicting unemployment and skill change—arguing for ensemble, task-level, and scenario-based exposure indicators to guide investment and social protection (Frank et al., 2023).
Inequality and redistribution: Compression of wage premia, polarization of high-skill labor markets, and sectoral vacancy-labor supply mismatches indicate a need for social insurance experimentation and adaptive, responsive retraining subsidies (Schaal, 15 Oct 2025, Marguerit, 24 Mar 2025).

7. Open Research Challenges

Despite substantial measurement advances, open questions remain:

Substitution vs. augmentation distinction at the task level is not always sharp; tasks with high exposure scores may also be most readily enhanced rather than supplanted, requiring longitudinal and causal identification of outcome heterogeneity (Colombo et al., 2024).
Adoption lags and technology diffusion: AI exposure does not always immediately translate into labor displacement; delays due to organizational adoption, regulatory hurdles, and societal acceptability must be integrated into forecasting models (Dominski et al., 11 Jul 2025, Meindl et al., 2021).
Dynamic task remapping: As tasks within occupations recompose in response to augmented capabilities, monitoring must encompass not just original task content, but also emergent patterns of AI–human complementarity (e.g., via activity-based applicability scores) (Tomlinson et al., 10 Jul 2025).
Heterogeneity in response: Firm-level strategies, team structures, local ecosystem resilience, and labor-market institutions may modulate exposure-outcome relationships, justifying a shift from static exposure mapping to simulated policy-intervention scenarios (Chopra et al., 29 Oct 2025).
Cross-national transferability: While most research centers on U.S. and U.K. task and labor-market structures, methodological adaptation to global occupational taxonomies, informality, and non-English task representation is an important ongoing agenda (Henseke et al., 30 Jul 2025).

Study of AI-exposed occupations now integrates multi-method task-based metrics, granular outcome tracking, and simulation-based projections. High exposure is a consistent feature in cognitively intensive, high-wage, analytic, and professional domains, with correlates in wage structure, employment volatility, and job design. Effective workforce and policy responses require dynamic monitoring, differentiated strategies by sector and region, and a long-term view of AI-driven labor transformation.