Mitchell, Ghosh, Passi: AI Agents Undermine Human Oversight

"AI agents pose significant risks as they are granted increasing autonomy." That is the opening line of a new position paper on arXiv by Margaret Mitchell, Avijit Ghosh and Samir Passi, submitted 24 August 2026 and classified under cs.AI and cs.HC.

The argument runs against the standard fix. The commonly proposed solution to agent autonomy is a "human in the loop", but the authors write that "current approaches to AI agent design impede effective human oversight". Worse, they claim, "the cognitive capacities required for it are also themselves degraded by extended use of AI systems." The people meant to be watching the machine lose the faculties needed to watch it.

The prescription is to raise oversight to the same status as capability. The paper says "a top priority in the advancement of AI agents should be supporting the situated goals and cognitive requirements of effective human oversight, treating the human needs of overseers at the same level of importance as AI agent capability." It draws on prior work in automation and human-computer interaction to outline design-level affordances and organizational protocols aimed at critical judgement and skill maintenance.

It is a position paper. The abstract publishes no experiments, no per-system measurements, and names no company whose agents are said to produce the effect. The closing line is a warning: without explicit support for the cognitive demands of oversight, AI agent systems "will continue to passively incentivize the degradation of the very human skills they rely on."