OpenAI study: 80% of workers will see AI change at least 10% of their job

A new paper from OpenAI researchers drops a number that should make everyone sit up: roughly 80% of the U.S. workforce could see at least one-tenth of their daily tasks altered by large language models like the ones powering ChatGPT. That’s not some distant sci-fi future. The study, authored by Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock, argues that these models are behaving like general-purpose technologies — the kind of foundational shift you get with electricity or the internet, not just a narrow tool for coders.

The researchers built a rubric to map GPT capabilities onto specific job tasks, blending human expert judgment with classifications generated by GPT-4 itself. The results are striking in their breadth. About 19% of workers may see at least half their work tasks impacted. And the disruption doesn’t discriminate by paycheck size — higher-income jobs might actually face greater exposure, not less. That flips the old automation script, where blue-collar and routine work typically got hit first.

Interestingly, the paper finds the impact isn’t confined to industries that were already riding a wave of productivity growth. It’s not just tech and finance. The tendrils reach into corners of the labor market that have been relatively stagnant, suggesting a reshuffling that’s both deep and wide. Daniel Rock, one of the co-authors from the University of Pennsylvania, helped ground the analysis in economic frameworks that treat these models as general-purpose, which raises the stakes considerably for policymakers.

What’s left unsaid is the timeline. The study maps potential exposure, not a forecast of job losses. A task being ‘affected’ could mean augmentation, not replacement. Still, the sheer scale — 80% is a number you can’t ignore — forces a conversation about retraining, wage pressure, and what it means to have a ‘high-skill’ job when a language model can draft your legal brief or analyze your spreadsheet in seconds. The question isn’t whether change is coming. It’s whether our institutions are built to absorb it.