AI Agent Protocol Risk Dashboard
Evaluating 15 agent protocols through Nadia Asparouhova’s Dangerous Protocols lens and Kafka Index. How much control does each protocol exert? How visible is that control? How costly is exit?
Stage 2
Stage 1
Stage 1
Stage 1
Stage 2
Stage 1
Stage 1
Stage 1
Stage 1
Stage 1
Stage 1
Stage 1
Stage 1
Stage 1
Stage 1
The payment, commerce, and token layers are the most dangerous. Infrastructure protocols (MCP, Goose, ACP, Agent Gateway) score well — they’re explicit, debuggable, and preserve human agency. But x402, AP2, UCP, and especially ElizaOS follow the pattern Asparouhova identifies: protocols designed to be invisible control you most completely. ElizaOS is the most Kafkaesque: token economics fuse financial identity with social belonging, making exit prohibitively expensive.
“Protocols are dangerous precisely because they control us so well. Though it may seem contradictory, the more powerful a protocol is, the harder it is to understand or explain it to others.” — Nadia Asparouhova, Dangerous Protocols