Published:
Readtime: 6 min
Every product is carefully selected by our editors and experts. If you buy from a link, we may earn a commission. Learn more. For more information on how we test products, click here.
Key points:
- Anthropic’s new study introduces “observed exposure,” measuring real-world AI usage rather than just theoretical capability, revealing that 97% of analysed tasks are now AI-feasible.
- Unlike previous automation waves, AI exposure is highest among higher-paid, college-educated, and female-dominated professional roles.
- High-exposure occupations have seen a 14% decline in entry-level hiring (ages 22–25) since late 2022, suggesting the “junior” workforce is being impacted first.
Is it too late to become a farmer or builder? That’s what workers might ask themselves after scoping Anthropic’s latest report, which details both the existing and forecasted impact of artificial intelligence (AI) on the US job market. The report’s primary aim was to highlight the types of professions that are “most exposed” to AI in the coming years, finding only “tentative evidence that hiring into those professions has slowed slightly for workers aged 22-25” at the present moment. Safe for now, but not for long, is one way you might summarise it.
While AI advancements have recently spurred massive layoffs at certain workplaces—Jack Dorsey’s Block, for example—the technology is not yet the disruptor that one might think it is based on headlines alone. In fact, Anthropic’s report found that “the unemployment rate for young workers in the exposed occupations is flat.”
Which jobs are most exposed to artificial intelligence?
Nevertheless, virtually no job sector is completely safe, at least not according to Anthropic. To visualise its findings, the report charted and graphed certain professions in terms of their exposure to AI. Unsurprisingly, Computer & Maths jobs are at the top of the proverbial chopping block. More to the point, Anthropic determined that AI can theoretically perform 94% of the tasks related to Computer & Maths jobs.
The next job category that’s most exposed to AI is Office & Administration, which currently employs an estimated 19.3 million Americans. According to Anthropic, AI can or will soon be able to perform 90% of Office & Administration tasks. Following that is Business & Finance (80% exposed to AI), followed by Architecture & Engineering (72% exposed), followed by Legal (70% exposed).
Anthropic’s report also determined that professions across the arts and social sciences are potentially vulnerable to massive AI disruption. That is to say, jobs in both Arts, Design & Media, and Life & Social Sciences are 68% exposed. Hot on their heels are Sales jobs (65% exposed) and Education jobs (62% exposed).
Jobs most exposed to artificial intelligence
| Rank | Occupation | Observed Exposure | Key Automated Task |
| 1 | Computer Programmers | 74.5% | Writing, updating, and maintaining software code. |
| 2 | Customer Service Representatives | 70.1% | Handling routine inquiries and order processing. |
| 3 | Data Entry Keyers | 67.1% | Reading source documents and entering structured data. |
| 4 | Medical Record Specialists | 66.7% | Compiling, abstracting, and coding patient data. |
| 5 | Market Research Analysts | 64.8% | Preparing reports and translating data into insights. |
| 6 | Sales Reps (Wholesale/Mfg) | 62.8% | Prospecting and automating product demonstrations. |
| 7 | Financial & Investment Analysts | 57.2% | Analyzing financial data and preparing forecasts. |
| 8 | Software QA Analysts & Testers | 51.9% | Executing test scripts and identifying software errors. |
| 9 | Information Security Analysts | 48.6% | Performing risk assessments and monitoring security. |
| 10 | Computer User Support Specialists | 46.8% | Troubleshooting software and hardware issues. |
Which jobs are the most ‘AI-Proof’?
On the flip side of the results are Construction jobs, which are currently the least exposed to AI, with only 18% exposure. That’s followed by Building & Grounds (20% exposed), followed by Installation & Repair (22% exposed), followed by Farming & Fishing (25% exposed), followed by Food Preparation & Serving (also 25% exposed). As a general takeaway, the more a job requires physical labor, the less likely it is to be impacted by AI in the near future.
To create its report, Anthropic assessed both the current and theoretical capabilities of Large Language Models (LLMs). Using that information, they created a unique graph that determines “Theoretical capability and observed exposure by occupational category.” The red portion denotes work that’s already being performed by Claude (Anthropic’s LLM), while the blue portion denotes the work that Claude will hypothetically perform as adoption grows.
Again, Anthropic’s report is a projection, meaning it’s built around hypotheticals. Regarding the current job market, the company determined that recent spikes in unemployment were more likely caused by other factors, including COVID-19 and international trade policies. It goes on to predict that the impacts of AI “might be less like COVID and more like the internet or trade with China,” making it harder to assess based purely on aggregate unemployment data.
| Rank | Occupation | Reason for Low Exposure |
| 1 | Cooks / Chefs | Physical food prep and sensory judgment (taste/smell). |
| 2 | Automotive Mechanics | Complex physical diagnostics and manual tool usage. |
| 3 | Groundskeepers / Landscapers | Physical labor in non-standardized outdoor environments. |
| 4 | Lifeguards / Ski Patrol | Real-time physical rescue and environmental monitoring. |
| 5 | Bartenders | Social-physical hybrid (mixing drinks + human interaction). |
| 6 | Plumbers | Navigation of unique, non-digital physical infrastructure. |
| 7 | Electricians | High-stakes manual repair and physical installation. |
| 8 | Nursing Assistants | Tactile patient care and physical mobility assistance. |
| 9 | Construction Laborers | Heavy manual labor and operation of physical machinery. |
| 10 | Carpenters | Precise manual shaping and assembly of materials. |
Whether Anthropic’s report is bleak, positive, or a little bit of both may depend on the individual. Some may fear massive job losses, while others will point to previous revolutions in history, whereby advancements in production and technology created entirely new modes of wealth, employment, and creativity. Along similarly optimistic lines, some experts predict that AI will make survival easier at large, resulting in programs such as Universal Basic Income (UBI) and reduced work hours.
Ethical discussions aside, Anthropic’s report effectively concludes that no desk job is safe should AI continue to advance. Even if employment hasn’t been seriously affected just yet—at least not in a quantifiable way—there’s no harm in picking up a trade or two in the meantime. It could one day make you a very rich person or at least give you consistent job stability. Hypothetically, at least.
Common questions about the impact of AI on the job market
According to the research, computer programmers (74.5%) and customer service representatives (70.1%) have the highest observed exposure. Other highly vulnerable roles include data entry keyers and medical record specialists, as their core responsibilities involve processing digital information that modern models can now manage with high proficiency.
These industries remain resilient because they require physical manipulation of unpredictable environments and manual dexterity that digital models cannot replicate. With exposure levels as low as 18%, occupations such as building, gardening, and electrical work are protected by the requirement for a physical presence and sensory-based problem solving.
While the overall unemployment rate has not spiked, there has been a 14% reduction in entry-level hiring for workers aged 22–25 in high-exposure sectors in the US. This suggests that businesses are beginning to use AI to complete junior-level tasks, such as basic research and administrative support, rather than hiring new staff for those specific roles.
Comments
We love hearing from you. or to leave a comment.