Anthropic Identifies the Jobs Most Exposed to AI Risks—Is Your Occupation Affected?

The workers most at risk from AI aren’t those earning the lowest wages, but are among the best-paid, best-educated professionals in the U.S., a new study from Anthropic has found.

Unlike past waves of automation, which hit blue-collar workers hardest, AI appears to be targeting so-called “knowledge work,” and the people who do it for a living may be more vulnerable than cashiers and cooks who earn far less than them. Almost three-quarters of a computer programmer’s core job tasks, for instance, are already being handled by AI—not someday, but in how it’s used today.

The study arrives as many Americans worry that AI automation may cost them their jobs, even as most research has been theoretical about which tasks AI might handle, with the most robust studies thus far showing limited effects on employment. Anthropic, the company behind the Claude AI model, is among the few organizations with direct access to real-world AI usage data at scale, making this among the more important studies on AI’s effect on jobs to date.

Which Jobs Face the Highest Real-World Risk

Anthropic’s study ranks computer programmers first, with AI being observed covering about three-quarters (74.5%) of their tasks. Customer service representatives follow at 70.1%, driven largely by AI handling customer inquiries through company APIs. Data entry keyers (67.1%), medical records specialists (66.7%), and financial and investment analysts (57.2%) round out the most exposed occupations.

Most striking about this list is that these aren’t the low-skill roles policymakers have historically worried about most in discussions about automation. Workers in the most exposed occupations earn about 47% more than workers in jobs with zero AI exposure, are far more likely to hold graduate degrees—17.4% vs. 4.5% in the zero exposure group—and are 16 percentage points more likely to be female. That’s because AI is targeting knowledge work, the kind involving analysis and writing.

AI’s Potential Reach vs. AI Usage Today

Many studies on the risks to jobs from AI have addressed what’s really merely a theoretical question: what tasks might AI take over from workers? Anthropic researchers Maxim Massenkoff and Peter McCrory took a different approach, creating a measure called “observed exposure” to track which tasks Claude is automating in real-world, work-related settings.

Their measure gives more weight to cases where AI replaces a worker’s output than to cases where AI simply assists them, giving a better sense of which workers are most likely to be replaced.

The Gap Between What AI Could Do and What It’s Actually Doing

Most AI-related automation remains theoretical, even as many, including Anthropic CEO Dario Amodei, are convinced that AI will replace jobs up and down the corporate ladder, rendering many jobs obsolete.

According to the Anthropic study, AI could perform 94% of the tasks in computer and math occupations. But, in practice, Claude is covering only about a third of those tasks. Legal work has often been mentioned as particularly vulnerable to AI, but Massenkoff and McCrory found relatively little automation thus far in that field. But those in the legal, computer and math-based professions shouldn’t break out the champagne just yet: This could mean, as Anthropic’s researchers put it, that “AI is far from reaching its theoretical capabilities.”

As AI’s deployment in the economy broadens, that gap should narrow, they argue, and so will job security in the occupations where it does.

What the Data Says About Jobs Lost—So Far

Mass layoffs from AI have yet to hit the most exposed workers identified by Anthropic. In line with previous research from the Federal Reserve Bank of Dallas and the Yale Budget Lab, Massenkoff and McCrory’s analysis of federal Current Population Survey data found no meaningful increase in unemployment rates for these workers since ChatGPT launched in late 2022.

But many have pointed out that worrying signs for the future of the labor market could be at its entry point. Among workers ages 22 to 25, the rate of starting a new job in a high-AI-exposure occupation has fallen about 14% compared with 2022. Workers over 25 show no such pattern, the study found. But Massenkoff and McCrory said there might be other reasons for these changes among the younger cohort of workers, noting that “the young workers who are not hired may be remaining at their existing jobs, taking different jobs or returning to school.”

For now, computer programmers, customer service workers, financial analysts and medical record specialists are the occupations to watch—not because AI-caused mass layoffs are imminent, but because the data has already started to move.