The 4 runtime detectors

Four threat families, detected at runtime on MCP and A2A traffic.

Submit my system

🧪

Indirect injection

Content collected or returned by a tool diverts the agent from its mandate. Detected on the incoming MCP flow.

⚡

Dangerous actuation

A side-effect action that does not match the mandate (write, call, transfer). Blocked before execution.

🔕

Alarm masking

A manipulation that gets an incident classified as benign. An attacker who gets their own intrusion approved turns your agent into the instrument of the attack.

📤

Exfiltration

A data leak through agentic channels (tools, responses, A2A delegation). Detected and cut fail-closed.

Measure your exposure

Submit your system, we quantify exposure across these four families.

Submit my system