Researchers have surveyed the Model Context Protocol (MCP) ecosystem, finding that AI agents are increasingly modifying external state. This shift in how agents interact with public blockchains through MCP, skills, and tool calling changes the threat model for attacks. The researchers identify four properties of blockchain execution layers that amplify the impact of attacks, making recoverable failures irreversible. They propose an attack-surface taxonomy and a Web3 risk-mapping matrix to categorize and mitigate these risks. --- Why it matters: This matters to AI engineers because it highlights the need for reevaluating security assumptions in agent interactions with public blockchains. The amplified impact of attacks due to blockchain properties requires more effective defenses, which this research aims to provide through a taxonomy and risk-mapping matrix. Source: https://arxiv.org/abs/2608.17275