Phishing attacks in blockchain (Ethereum) create a notable security vulnerability for users to get exploited by fake actors who are trying to trick users into divulging sensitive information frequently resulting in substantial financial losses. Conventionally, phishing detection methods rely only on centralized systems, which fall short in terms of scalability and breach the characteristic nature of blockchain technology like privacy-preserving and decentralized approach. In this paper, we propose the use of Federated Learning (FL) for phishing detection in Ethereum which is a decentralized mechanism. By implementing an FL approach, the proposed model can allow multiple clients to train a global model collaboratively by using locally stored data. It not only ensures preserving the privacy of sensitive data but also enhances scalability. This model uses weighted federated averaging which results in an accuracy of 79.96% and F1-score of 79.94% in detecting phishing attacks. All this is achieved while preserving privacy and making it an acceptable solution for dynamic, decentralized environments like Ethereum. This work exemplifies the perspective of FL in elevating the security and privacy of blockchain networks.