AI is reshaping the labor market. Here's what the data actually shows.
Anthropic's new research measures real AI displacement risk across occupations. No mass unemployment yet, but hiring is slowing for young workers in exposed roles. What this means for international professionals in Europe.
The short version#
Anthropic published a new study on March 5, 2026 measuring how AI is actually affecting jobs. The researchers combined theoretical AI capability with real-world usage data from Claude to create a new metric called "observed exposure." The headline: no systematic increase in unemployment for AI-exposed workers yet. But hiring of young workers (ages 22 to 25) into exposed occupations has slowed by about 14%.
For international professionals navigating European labor markets, this matters. A lot.
What the study found#
Maxim Massenkoff and Peter McCrory from Anthropic introduce a measure that goes beyond theoretical guesses about which jobs AI could replace. Instead, they look at which tasks AI is actually performing in professional settings. The difference is significant.
The gap between theory and reality is large. For Computer and Math occupations, 94% of tasks are theoretically doable by LLMs. But actual coverage from real usage data is only 33%. AI can do far more than it currently does. The question is how fast that gap closes.
The most exposed occupations are:
- Computer Programmers (75% task coverage)
- Customer Service Representatives
- Data Entry Keyers (67% task coverage)
- Financial Analysts
- Technical Writers
30% of workers have zero exposure. These are roles like cooks, bartenders, motorcycle mechanics, lifeguards. Their tasks don't appear in AI usage data at all.
Workers in exposed occupations tend to be:
- More educated (graduate degree holders are 4x more represented)
- Higher-paid (47% more than unexposed workers on average)
- More likely to be female (16 percentage points more)
- More likely to be white or Asian
No unemployment spike, but a hiring signal. The researchers compared unemployment trends for high-exposure and low-exposure workers since late 2022. No meaningful difference. However, young workers (22 to 25) are starting new jobs in exposed occupations at roughly half a percentage point lower rate. Job finding rates in low-exposure occupations remain stable.
Why this matters for international talent#
Most international professionals in Finland and across Europe work in exactly the roles this study flags as highly exposed: software development, data analysis, customer service, financial services, technical writing.
These are the fields where skilled immigrants have traditionally found their path into European labor markets. They're knowledge work roles that don't require perfect local language skills and rely on transferable, technical expertise.
Three things stand out.
1. The hiring slowdown hits newcomers first.
The study's most concerning finding isn't about layoffs. It's about reduced hiring. Young workers, career changers, and labor market entrants are the first to feel the squeeze. For international professionals, who are almost always labor market entrants in their new country, this is directly relevant.
If companies are hiring fewer people into customer service, data entry, and financial analysis roles because AI handles parts of those jobs, the people who lose out first are those without local networks, local references, and incumbency advantages. That means immigrants.
2. High-skill roles are more exposed, not less.
The common assumption is that AI threatens low-skill work. This study shows the opposite. Higher education and higher salaries correlate with greater exposure. The occupations most affected are the ones international professionals specifically train for and move countries to pursue.
This challenges the idea that getting another degree or certification is a safe strategy. The roles those credentials target are exactly where AI coverage is highest.
3. The gap between capability and adoption is temporary.
Right now, AI performs 33% of tasks it theoretically could in computer science roles. That number will grow. The study explicitly frames this as a tracking exercise, measuring how the gap closes over time.
For anyone planning a career in Europe over the next five to ten years, the gap narrowing means the labor market you enter might look different from the one you planned for.
What you can do#
This research doesn't predict mass unemployment. It shows a labor market that is shifting gradually, with the effects concentrated in specific roles and age groups. For international professionals, the takeaway is practical.
Learn to work with AI, not compete against it. The study distinguishes between automated use (AI replacing tasks) and augmented use (AI assisting with tasks). Augmented use receives half the weight in their exposure metric. Workers who use AI as a tool, rather than doing tasks AI can fully automate, are better positioned.
Don't rely on a single technical skill. Computer programming has 75% task coverage. But that doesn't mean all programming jobs disappear. It means the nature of the work changes. Programmers who can architect systems, manage teams, and translate business needs into technical solutions do work that AI doesn't cover.
Move fast. The hiring slowdown for young workers is a signal, not a catastrophe. The labor market still works. But the window for certain entry-level paths may be narrowing. Getting established in a role now, before coverage increases further, gives you incumbency advantages that help later.
Watch the data. The researchers plan to update this analysis as new usage data comes in. The task and occupation-level exposure data is publicly available on HuggingFace. Understanding which tasks in your field are being automated helps you focus on the ones that aren't.
The bottom line#
AI is not causing mass unemployment. Not yet. But it is quietly changing who gets hired and for what. The workers most affected are educated, higher-paid, and disproportionately in roles that international professionals depend on.
For the Herizon community, where over 54% of tracked members are employed and many work in exactly these fields, this study is worth reading closely. The labor market is not collapsing. But it is moving. And staying ahead of the shift matters more than waiting for it to become obvious.
Source: Massenkoff, M. and McCrory, P. (2026). "Labor market impacts of AI: A new measure and early evidence." Anthropic. Full paper
