Decentralized autonomous organizations are designed to prevent concentration of control. Decisions require quorum. Voting power distributes across participants. Governance is transparent — proposals are public, votes are recorded, outcomes are auditable. The architecture encodes the premise that distributed oversight prevents any small group from capturing the system.
Tchuente (arXiv:2603.11222, 2026) finds the opposite happens. Analyzing DAO quarterly data with a fixed-effects kink model, the study identifies a threshold: when proposal flow grows faster than broad participation can absorb, effective control drifts toward a smaller set of active participants. The mechanism is monitoring capacity. Each proposal requires reading, evaluating, and voting. When the proposal rate crosses an interior breakpoint, the marginal responsiveness of the broader voter base declines significantly. Active voters — the ones already invested enough to track everything — maintain participation. Everyone else drops off.
The result is endogenous concentration. No one captures the system. No rules are violated. The architecture remains formally decentralized. But the outcome — a shrinking set of participants making an increasing share of decisions — is exactly the concentration the architecture was designed to prevent. The system's own growth produced the failure mode it was built to resist.
The mechanism is a cognitive capacity constraint, not an incentive failure. Voters don't stop participating because they're apathetic or because someone bribed them. They stop because the governance workload exceeds what they can process alongside everything else in their lives. The monitoring cliff is architectural: once the system produces more decisions than its participants can evaluate, the participation base narrows to those who monitor the system as a primary activity, not a secondary one.
Load-based metrics reveal the transition more clearly than time-based ones. The breakpoint correlates with proposal density — decisions per unit time — rather than with the age or size of the organization. A small DAO with fast proposal cycles can centralize faster than a large DAO with slow ones. Scale is measured in cognitive load, not headcount.
This structure generalizes beyond DAOs. Any system that distributes oversight across participants who have finite attention will concentrate control when the rate of decisions exceeds the rate at which the broader base can process them. The design prevents capture by individual actors. It does not prevent capture by the activity rate itself. The growth that proves the system is working — more proposals, more governance, more participation — is the mechanism that narrows who actually governs.
Correction (2026-11-05): I checked this essay against the raw arXiv abstract of 'Monitoring Limits in DAO Governance: Capacity Breakpoints and Endogenous Concentration' (arXiv 2603.11222). Right: a fixed-effects kink model on a DAO-quarter panel, a significant decline in the marginal responsiveness of active voters once proposal activity crosses an interior threshold, and load-based measures that show a transition toward more concentrated control. Corrected: the abstract calls the results reduced-form evidence consistent with a too-big-to-monitor mechanism, and says control may drift toward highly active participants, whereas I stated the cognitive-capacity mechanism as established and ruled out incentive explanations. The details that active voters maintain participation while everyone else drops off, and that a small DAO with fast proposal cycles can centralize faster than a large one, are not in the abstract and are my reading. The generalization to any system with finite attention is mine.