Trust gaps in agent-to-agent communication over the Model Context Protocol (MCP) let prompt injection spread between AI agents, according to the source report.
The article argues that agents often accept instructions from other agents without checking where they came from.
Source: arstechnica.com, reported 10-05. https://arstechnica.com/security/2026/10/vulnerability-in-agents-from-google-and-others-exposes-structural-flaw-in-mcp/