OpenAI published new economic research on 16 September 2026 arguing that workers are using AI to take on work outside their traditional job boundaries S¹. The claim pushes against the common framing of workplace AI as a tool that simply speeds up tasks people already do, suggesting instead that the technology is generating new recurring activities that did not exist in people's job descriptions before. What has not been checked: OpenAI has released no error bars, no study code, and the assertion rests on the vendor's own summary rather than an independently audited dataset S¹.
My read: I'd treat this as a directional signal from a company with a clear commercial interest in the outcome. The phrase "beyond traditional roles" is doing a lot of heavy lifting here, and until the full report is public, I would hold off on treating any of this as settled.
The research is part of OpenAI Economic Research's "Work at the Frontier" series, begun in July 2026 to study employees who pick up duties that their official job titles do not cover P². That earlier page frames the series as an investigation into staff taking on responsibilities that fall outside their formal job descriptions P². The September publication appears to be the latest instalment in that ongoing line of inquiry, though OpenAI has not explicitly labelled it as such on the page itself.
Separately, OpenAI has released GABRIEL, an open-source toolkit on GitHub that lets social scientists and data scientists measure quantitative attributes in text, images, or audio using the GPT API P³. The repository, created on 5 February 2026, carries an Apache 2.0 licence and lists 418 stars and 67 forks. Its top contributor is listed as hemanth-asirvatham P³. OpenAI has not stated whether GABRIEL was used in this specific research.
The summary omits adoption rates, industries, and productivity metrics
The published page describes workers using AI beyond traditional roles and identifies new activities that become recurring parts of their work S¹. What it does not contain is specific adoption rates, named industries or job titles, productivity measurements, or any comparison to prior periods. The evidence pack confirms these gaps explicitly. A reader looking for hard numbers on, say, what share of marketers now do data analysis because of AI will not find them here.
For a workforce planner at a mid-sized firm, nothing on this page changes what lands on their desk: there are no adoption rates to fold into a staffing model, no industry breakdowns to benchmark against, and no productivity figures to justify a reorganisation S¹. The one concrete tool available right now is GABRIEL, which a research-minded analyst could use to code and measure qualitative data such as interview transcripts or open-ended survey responses about how staff actually spend their time P³.
OpenAI has not announced a date for the full report's release.
Sources: S1 — How workers are unlocking new ways of working · P2 — How AI is expanding what people do at work | OpenAI · P3 — openai/GABRIEL
Written from 3 sourced items, 3 of them primary.
