Muti-block withholding attack

Ethereum 2.0\u2019s adoption of the Gasper Proof-of-Stake protocol improves scalability and energy efficiency but also exposes new security risks. This paper introduces a novel multi-block withholding attack that exploits Gasper\u2019s incentive mechanism to maximize attacker rewards through strategic block delays. We construct a fine-grained MDP model and apply deep Q-learning to derive near-optimal attack strategies, effectively addressing challenges of chain ambiguity and local predictability.  To facilitate further research, we release a dataset containing simulation traces and attack outcomes related to Ethereum 2.0\u2019s Gasper protocol.