More than three years after the launch of ChatGPT, the influence of artificial intelligence (AI) on the job market is still difficult to detect in broad employment statistics. Yet early signs of disruption are beginning to appear, especially in entry‑level positions within sectors most vulnerable to automation. A new joint study from Coface and the Observatory of Threatened and Emerging Jobs (OEM) provides a detailed mapping of how AI‑driven automation affects specific tasks across different occupations, revealing a clear shift in the boundaries of what can be automated.
Unlike previous waves focused on routine work, AI is increasingly targeting cognitive, complex, and highly skilled tasks, raising the risk of major changes in how work is organized.
A New Method to Measure AI Automation Risks
The study offers an in‑depth mapping of the domains where artificial intelligence (AI) is most likely to reshape work. This granular approach uncovers vulnerabilities that broad labor statistics often overlook, as exposure to AI varies widely across tasks, occupations, sectors, countries, and regions.
The OEM’s methodology tackles three major shortcomings found in many current assessments: insufficient detail in occupational analysis, low reproducibility in expert‑based evaluations or AI‑generated judgments, and the lack of a truly forward‑looking perspective on the different stages of AI development.
The analysis breaks down 923 professions into tasks, which are further divided into elementary actions described as triplets (verb, object, context). This structure enables a far more precise evaluation of how exposed each task is to AI‑driven automation. These elementary actions are then scored using transparent, reproducible rules.
This approach directly addresses the previously identified gaps. It sharpens occupational analysis by rating generic basic actions independently of job titles, enhances reproducibility through explicit and auditable scoring rules, and introduces a forward‑looking dimension by projecting task exposure across multiple phases of AI development — five phases in this study — rather than offering a single static snapshot.
Working with the OEM, Coface has strengthened the framework by developing a method to weight tasks based on their importance and frequency, refining future‑oriented scenarios and scoring rules, and expanding the empirical scope to nearly thirty countries.
The evaluation focuses solely on the technical exposure of tasks to automation and does not attempt to estimate net job losses. By design, it does not incorporate demand‑side dynamics, the emergence of new tasks, or the practical frictions that may slow or limit real‑world AI adoption.
Varying exposure across occupational groups: AI primarily targets cognitive and information-related activities
The study reveals a significant shift compared to previous waves of automation. Unlike robotics or traditional software, AI primarily targets complex, non‑repetitive cognitive tasks. Its impact is highly uneven: it begins at the task level and then cascades differently across occupations, occupational groups, and the sectors where these roles are concentrated.
In the main scenario — focused on the deployment of agent‑based AI — around one in eight occupations surpasses the 30% threshold of automatable tasks. Crossing this threshold signals a profound transformation of the profession and may lead to substantial workforce redeployment, though not necessarily the disappearance of the occupation.
The most exposed roles are concentrated in highly cognitive, information‑intensive fields such as engineering, IT, administrative functions, finance, law, and certain creative or analytical professions.
Number of professions with ≥ 30% of tasks that can be automated, by occupational group, “Special Agent” scenario
Data for graph in .xlsx format
Conversely, the least exposed occupations are mainly based on manual work or direct human interaction, which remains difficult to standardize or automate. These include manufacturing, construction, maintenance, transport, catering, cleaning, and certain care and support services.
The study also evaluates how much actual work content is at risk in each labor market by comparing the share of automatable tasks in each of the 923 occupations with employment levels. By grouping occupations into eight broad categories, it identifies which professional groups face the highest exposure to AI‑driven automation.
The results are clear: more than 25% of work content could be automated in management and administration, creative professions, law and finance, as well as engineering and IT. By contrast, face‑to‑face services, technical roles, crafts, and industrial production remain below the 10% threshold. Jobs in care, education, sales, and other people‑focused professions fall in between — some tasks are exposed, but the human dimension continues to provide strong protection.
Significant disparities between countries
The analysis shows wide differences in national exposure to AI automation, ranging from around 12% of work content at risk in Turkey to nearly 20% in the United Kingdom. These gaps are largely driven by economic structure, which shapes employment patterns and determines how many tasks can realistically be automated.
Advanced economies with a strong focus on cognitive and knowledge‑based services appear most exposed. Alongside the UK, countries such as the Netherlands, Ireland, and Luxembourg have a high concentration of information‑intensive jobs. In contrast, economies more focused on trade, personal services, construction, transport, or physically demanding work show lower exposure levels. Overall, the study identifies five broad country groups with similar automation profiles.
Beyond employment: value sharing, social protection, education, new dependencies… many questions currently without answers.
The impact of AI extends well beyond employment alone. Because it increasingly affects skilled and well‑paid professions, AI adoption could alter long‑established economic and social balances.
By automating certain tasks in high‑skilled roles, AI may shift a growing share of value creation from labor to capital. For countries that rely heavily on labor taxation, this could create a double fiscal challenge: lower tax revenues combined with higher public spending on unemployment support and workforce reskilling.
The findings also raise questions about education and qualifications. If tasks linked to long academic paths become easier to automate, the connection between education, wages, and job security may weaken. While higher education remains essential, employers may place more value on complementary skills such as judgment, adaptability, and the ability to supervise and use AI effectively.
In addition, AI development could create new geopolitical and operational risks due to the concentration of key assets — such as semiconductors, large language models, and data centers — within a small number of firms and countries.
Conclusion: a transformation capable of reshaping work
While the pace and scale of these changes remain uncertain, and technical exposure does not automatically translate into job losses, one conclusion is clear.
AI is no longer confined to the margins of the labor market. It increasingly affects cognitive, non‑routine, and skilled activities that were long considered secure. Because these roles are central to income generation, productivity, and tax revenues, the spread of AI is likely to reshape jobs, value creation, and economic balances in lasting ways.
> Download the full study (.pdf) or watch the authors' keynote given as part of the Coface Country Risk Conference
