Ethics
OpenAI's Own Research Admits Workers Aren't Benefiting From AI Productivity Gains
A quietly published April 2026 policy paper from OpenAI acknowledges that productivity gains from AI tools are not flowing to the workers who use them — raising uncomfortable questions about the company's social contract.
By Patrick T ·
In April 2026, OpenAI published a policy paper that received far less attention than it deserved. Buried within a broader discussion of AI's economic effects was a frank acknowledgement: workers who use AI tools may agree that their productivity is rising, but they do not believe they are receiving the benefit of that productivity gain. The paper cited survey data showing that a majority of AI-using workers felt their employers — not themselves — were capturing the value created by their increased output.
The timing is striking. OpenAI published this finding in the same month it launched DeployCo, a $4 billion consulting subsidiary explicitly designed to increase productivity inside client organisations. The company is, in effect, acknowledging a problem with one hand while building a business that accelerates it with the other.
The Productivity-Benefit Gap
The phenomenon the paper describes is not new to economists. It mirrors the pattern observed during the introduction of enterprise software in the 1990s and 2000s: productivity gains were real, but they accrued primarily to shareholders and senior management rather than to the workers whose labour became more efficient. The difference with AI is speed and scale. Enterprise software took decades to permeate the workforce. AI tools are being deployed to millions of workers simultaneously, and the productivity gains are measurable within months.
The KPMG and PwC deployments illustrate the dynamic concretely. PwC reported that Claude reduced insurance underwriting time from ten weeks to ten days — a 93% reduction in time-to-completion for a specific workflow. The professionals who previously spent ten weeks on that task now complete it in ten days. The question the OpenAI paper implicitly raises is: what happens to those professionals' workloads, compensation, and job security as a result? If the answer is 'they now handle seven times as many clients for the same salary,' the productivity gain has been captured entirely by the firm.
The Structural Problem
OpenAI's paper stops short of proposing solutions, which is itself revealing. The company acknowledges the problem but frames it as a policy challenge for governments rather than a responsibility for AI developers. That framing is convenient but contestable. AI developers make choices about how their tools are designed, priced, and deployed. A model that is sold exclusively to employers, with no mechanism for workers to negotiate the terms of its use, is a design choice — not a neutral technical fact.
Some economists and labour advocates are beginning to argue for 'AI dividends' — mechanisms by which workers who contribute to AI productivity gains receive a share of those gains, either through profit-sharing, reduced working hours, or direct payments. The concept is nascent and faces significant political and practical obstacles. But the fact that OpenAI's own research is surfacing the productivity-benefit gap suggests that the question is no longer hypothetical. It is a live problem, and the AI industry's response to it will define its social legacy as much as any benchmark score.
Topics: OpenAI, AI Ethics, Labor, Productivity, Economic Impact