The proliferation of AI-based detectors, particularly those employing graph analytics, poses a substantial challenge to the viability of covert communication over blockchain networks. These AI methods are capable of identifying subtle structural and statistical anomalies, thus exposing most existing covert communication channels. Moreover, current blockchain-based covert communication schemes suffer from limited information filtering efficiency and a higher risk of identity disclosure, which jointly undermine the security and effectiveness of the communication process. To address these challenges and effectively evade AI-based detection while improving overall communication security and efficiency, this paper designs a hybrid blockchain covert communication method based on Ethereum transaction amount and address. We encode the hidden information as the Ethereum transaction amount and send transactions using dynamic special transaction addresses created by the Hash-Based Message Authentication Code (HMAC) algorithm. The encoding rules are determined by the varying CPU models and their corresponding running times of the Verifiable Delay Function (VDF). This dual approach neutralizes both feature-based and graph-based detection by making the transactions statistically unremarkable and structurally disjointed on-chain. Implementation on the Ethereum testchain demonstrates that our method provides a highly secure and efficient channel, capable of successfully evading advanced AI surveillance.