Researchers say expanding artificial intelligence (AI) could weaken workers’ bargaining power and citizens’ influence if competition pushes companies and governments to delegate more decisions to machines, even without a sudden crisis.
A 2025 paper by Jan Kulveit and five coauthors calls this possible pattern “gradual disempowerment.” They argue that institutions may benefit from assigning work to AI while becoming less dependent on the people whose labor and participation give them influence. The paper presents a scenario, not evidence of a societywide loss of control, and gives no probability or timetable.
In the authors’ account, companies that delegate effectively could gain an advantage over rivals. Workers could lose income and bargaining power as a result. The authors also suggest that governments relying less on people’s work and participation could become less responsive to them.
They argue that changes in economic power could affect culture and public policy, reinforcing shifts across those areas. As an analogy for economic power extending beyond a market, the paper points to the tobacco industry’s influence on politics and culture; it does not present that history as an AI precedent.
Present-day workplace evidence is narrower. An April 2026 report by the Massachusetts Institute of Technology (MIT) Working Group on Generative AI and the Work of the Future studied more than 20 companies across four industry groups. In the AI applications examined, workers took on more supervision of automated systems while manual process operation accounted for fewer duties. The report found no consistent pattern in how those applications changed skill requirements.
An August 2026 position paper by Margaret Mitchell, Avijit Ghosh and Samir Passi raises two distinct oversight concerns. Repeated requests to approve an AI agent’s actions may cause users to pay less attention, they argue, while extended automation may erode the skills needed to supervise such systems.
Kulveit and his coauthors propose tracking AI’s share of economic output, its role in major corporate decisions, the cultural content it produces and its use in policy and legal processes. They also discuss human oversight requirements, limits on AI autonomy in selected areas and taxes on AI-generated revenue. The authors say such measures could sacrifice potential value and invite circumvention, and may be less effective without broad adoption and international coordination. The paper leaves open where to set thresholds for action and how to assess whether interventions work.
