Artificial intelligence reshaping office work is no longer just a hypothetical, but pinning down exactly how and when that disruption lands has been surprisingly difficult for economists. Anthropic is now trying to get ahead of that problem. The company says it has built a system specifically designed to track how vulnerable different jobs are to AI automation before the damage shows up in unemployment data. The goal is to catch the early signals, not just document the aftermath.
How Anthropic’s Exposure Index Actually Works
According to a new research paper by Anthropic economists Maxim Massenkoff and Peter McCrory, the system works as an index that measures how exposed different occupations are to automation by large language models. Rather than looking at job titles broadly, it breaks roles down into the individual tasks they involve. If a significant portion of those tasks can be handled by AI and are already being handled by AI tools in practice that occupation scores higher on the exposure scale. It’s a more granular approach than most previous attempts at measuring this kind of risk.
The Jobs Most At Risk Right Now
The findings are uncomfortable reading if you work in tech. Programmers rank among the most exposed occupations, with AI systems reportedly capable of handling around three-quarters of their common tasks. Customer service representatives, data entry workers and medical records specialists also sit high on the list, roles where a large chunk of the day-to-day work involves the kind of structured, repeatable tasks that language models handle well. On the other end of the spectrum, jobs that require physical presence and hands-on work appear far more insulated. Cooks, lifeguards and dishwashers, for instance, remain largely outside the reach of current AI capabilities, for now, at least.
Has The Disruption Actually Started?
Here’s where the research gets nuanced. Despite the growing anxiety around AI-driven job losses, the study doesn’t find strong evidence that mass displacement has begun. Unemployment rates between AI-exposed and AI-resistant occupations haven’t diverged dramatically since ChatGPT arrived. But there are early tremors worth watching. Hiring for young workers aged 22 to 25 appears to be slowing in AI-exposed roles, a pattern that researchers say tends to show up in occupations where AI tools can substitute for entry-level work. It’s not a crisis yet. But it’s not nothing either.
