Why Ethereum Needs Fairness Mechanisms that Do Not Depend on Participant Altruism

This collection of recent research highlights systemic risks and attack surfaces across decentralized systems, model serving, and edge/mobile AI. Key findings: (1) Ethereum proposer centralization is severe—~91% blind-delegate to external builders and <1.4% consistently act to preserve decentralization/censorship-resistance—implying that fairness mechanisms that assume altruistic proposers are ineffective and that incentive/penalty mechanisms are needed to mitigate censorship and centralization risks. (2) LLM and MoE serving (dense and sparse) expose availability and latency risks—expert load-