Agentic AI could enclose users, warns study on MCP governance

A new academic paper, "Agentic AI: User Empowerment or Enclosure?" (arXiv:2608.06510), argues that the promise of agentic AI—systems that act on users' behalf, from filtering content to negotiating prices—is not preordained by technology alone. Authors David Gamba, Daniel M. Romero, and Grant Schoenebeck conduct a comparative case analysis of four mature domains where similar agency emerged: browser-based ad blockers, platform recommender systems, financial robo-advisors, and email spam governance. Their central finding: decisions about whose interests these agents serve were resolved through technical choices—API configurations, protocol governance, industry standards, and defaults—which were fundamentally political, though often framed as neutral engineering.

This "depoliticization" has a critical consequence: individual outcomes and collective contestation capacity can move in opposite directions. For example, spam inbox quality improved dramatically, yet the organized capacity to contest spam governance collapsed. Where intermediary institutions sustained adversarial challenge, user-aligned agency proved more durable; where proprietary infrastructure and closed standard-setting absorbed contestation, displacement compounded. Applying this lens to agentic AI, the authors identify worrying parallels: governance arrangements are consolidating around the Model Context Protocol (MCP) and the Agentic AI Foundation. These standards are settling configurations that define what agents can do—before users or the public can meaningfully shape them. The paper concludes that without deliberate intervention in these technical-governance layers, agentic AI may empower individual users in small ways while enclosing their collective agency at a structural level.