Asynchronous consensus mechanisms offer notable advantages, such as resistance to censorship and high fault tolerance, making them highly suitable for high-concurrency applications, including IoT and supply chain finance. However, existing research has focused primarily on performance and efficiency improvements, while privacy protection capabilities remain inadequate. Although privacy-preserving techniques like ring signatures and homomorphic encryption can be applied in simple transaction scenarios, they often fail in complex environments such as cross-chain auditing due to key leakage or excessive computational overhead. To address this issue, this paper proposes Shadow, an asynchronous Directed Acyclic Graph (DAG) consensus algorithm that reconstructs the network topology through parallel transaction verification. The approach introduces a Primary Anchor Layer (PAL) and a Dynamic Shard Validation Layer (DSVL), and incorporates zero-knowledge proofs to enable dynamic privacy switching within a hierarchical verification architecture. This achieves an optimal balance between privacy protection and consensus efficiency without compromising auditability. Experimental results demonstrate that in a 128-node environment, Shadow achieves a throughput of 16,000 TPS in transparent mode (without zero-knowledge proof), 1.5 times that of Hashgraph and 3.8 times that of DAG-rider. Moreover, it maintains a throughput of 11,500 TPS in privacy mode (with zero-knowledge proof), exceeding both baseline methods by 37.5% and 43.8%, respectively.