Goldman Sachs: AI Exposure Now Touches Two-Thirds of US Occupations

Two-thirds of US occupations are exposed to some degree of automation by artificial intelligence, according to Goldman Sachs Research, whose analysis of global labor markets remains one of the most cited benchmarks for sizing AI's impact on work.

The figure comes with an important qualifier: exposure is not displacement. In most cases, Goldman Sachs analysts find, AI exposes a share of an occupation's tasks rather than the occupation as a whole. Roughly a quarter to a half of the workload in affected occupations could be automated, which the bank frames as a partial shift in how work gets done rather than a wholesale elimination of roles.

The headline global number is larger. Applying a task-based methodology across major economies, Goldman Sachs Research estimates that around 300 million full-time jobs worldwide could be affected by automation driven by generative AI. The analysis draws on occupation-level task content data, mapping the capabilities of large language models and related systems onto the specific tasks each job contains.

What the numbers mean at the sector level

The exposure is unevenly distributed across the economy. Administrative support and legal occupations show some of the highest shares of tasks that AI could automate, according to the bank's task-mapping. Office and administrative roles, in particular, feature heavy concentrations of routine information processing — drafting, summarizing, data entry, scheduling — that overlaps directly with what current generative systems do well.

At the other end of the spectrum, physically intensive occupations such as construction, installation, maintenance and repair show minimal exposure. The work is hard to automate not because it is more skilled, but because it happens in unpredictable physical environments that current AI systems cannot reach.

What this means for an individual's market position

For a worker in a high-exposure occupation, the data points to a specific strategic conclusion: the tasks most likely to be automated are the routine, text-heavy and pattern-based portions of the job. The premium is shifting to the parts of the role that involve judgment, client relationships, negotiation and accountability — the components that do not transfer cleanly to a model.

For workers in low-exposure occupations, the message is different. Their positions are comparatively insulated from this wave of automation, but the Goldman Sachs analysis is explicit that the technology diffusing through adjacent white-collar functions will change the tools, expectations and productivity benchmarks around them.

The bank's labor-market framing also addresses the historical pattern. Past waves of automation displaced some jobs but were offset by the creation of new roles and occupations following major technological shifts. Goldman Sachs Research treats that pattern as a plausible reference case, while noting that the breadth of generative AI — cutting across nearly every white-collar sector at once — distinguishes it from earlier, narrower technologies.

The macro side: productivity, not just displacement

The same research estimates that widespread AI adoption could lift annual US labor productivity growth by roughly 1.5 percentage points over a decade, with smaller but positive effects across other major economies. That outcome depends on the technology being deployed to augment workers rather than substitute for them — a distinction the bank draws from how previous general-purpose technologies, from the electric motor to the personal computer, ultimately fed through to output.

The timing matters as much as the size. The productivity effects, in the bank's modeling, build gradually and depend on adoption, investment and complementary innovation in software and business processes. Workers and employers making plans now are operating ahead of the steepest part of the adoption curve.

Caveats worth weighing

Task-based estimates carry known limitations. They measure technical feasibility — what a system could do — rather than what firms will actually implement once cost, regulation, data quality and organizational friction enter the equation. Goldman Sachs' figures describe the outer boundary of exposure, not a forecast of layoffs.

Looking ahead, the bank's researchers indicate they will continue tracking adoption evidence — hiring patterns, productivity data and investment flows — to test whether the task-exposure estimates are converting into measurable labor-market outcomes, making the coming quarters of jobs data a live test of the 300-million-job framework.