Advances in artificial intelligence models have long promised to automate vast swathes of desk work, but the physical world is now squarely in the crosshairs. On September 30, 2026, Anthropic published a robot exposure index designed to evaluate just how well current machines can execute physical tasks in the wild. By mapping physical capabilities against actual job requirements, the researchers at Anthropic have established a stark baseline for where mechanical labor intersects with human employment—and the findings suggest blue-collar workers face a much steeper cliff than anticipated.
Mapping Physical Exposure Across Tasks
To construct this measure of job exposure, Anthropic deployed Claude to assess the competency of present-day robots in performing concrete work tasks. The analysis defined robots strictly as autonomous physical machines capable of sensing and acting upon their environment. According to the findings, robots can theoretically perform three-quarters of physical tasks in the United States, which translates to 34% of total working hours, though these capabilities largely remain confined to structured settings.
"Robots, which we define as autonomous physical machines that sense and act, can perform three-quarters of physical tasks in the US, making up 34% of working hours, but mostly in limited settings."
When combining physical exposure with cognitive automation via large language models, the numbers turn grim for the workforce: roughly 80% of all job tasks by working time are now exposed to either robots or LLMs. However, the distribution of this threat is wildly uneven. Driving and warehouse operations face immediate disruption from currently available hardware, while nursing and general repair remain heavily insulated because present-day machines can barely manage the unstructured chaos of those jobs, even in highly controlled environments. Demographically, the index shows that workers most exposed to robotic replacement are disproportionately male, less educated, and lower-paid.
Historical Trends and the Cost Barrier
This is not a sudden shock, but the continuation of a grinding trend. A backtest included in the research indicates that jobs with higher robot exposure experienced steeper declines in both wages and employment over the past 50 years. Over that same timeframe, robots have steadily expanded their capabilities by roughly 2% each year, quietly absorbing physical work they previously could not touch.
"Robots are cost-competitive for just 0.3% of job tasks.
Yet functional viability and market reality remain worlds apart. Economic friction continues to act as the ultimate bouncer for the robotics industry: robots are cost-competitive for a meager 0.3% of job tasks today. If hardware price declines follow historical trajectories, it will take another 40 years for that share to crawl to 10%.
Anthropic's index demonstrates that raw physical capability alone does not dictate immediate market displacement, as economic realities, regulatory hurdles, and operational constraints continue to slow down adoption. While combined task exposure hits the 80% mark when paired with LLMs, the research leans heavily on historical price curves and controlled operating environments. The real variable, and the one worth watching, is whether novel machine learning architectures can accelerate physical generalization and brutally compress that 40-year economic curve.