The increasing usage of AI agents drives the shift from individual agent-based systems toward multi-agent deployments. In this scenario, an open standard has emerged: Agent-to-Agent Protocol. Abbreviated A2A, it provides a communication layer between AI agents, ensuring interoperability regardless of their internal architecture or implementation details. However, since this protocol is relatively new, it also introduces practical security challenges, especially when it comes to the Identity and Access Management part. To address these challenges, this thesis proposes a security framework for multi-agent architectures, based on Decentralized Identifiers and Verifiable Credentials. Indeed, the main contribution of this thesis is an approach to enhance the authentication phase of the A2A protocol by embedding DIDs to enable a cryptographically verifiable identity and using VCs to represent the permissions held by agents..