MCP rug-pull attack worries mount

This article warns of “rug-pull” attacks against MCP-based AI agent tooling, where servers can silently change tool definitions after user approval, enabling post-deployment drift that can exfiltrate sensitive data or perform unauthorized actions; it explains the lack of hashing/versioning in MCP, notes the ease of exploitation, and recommends versioned snapshots, cryptographic manifests, comprehensive audit logs, and behavioral baselines to mitigate the risk.