What is ArmorIQ?

ArmorIQ is a security platform for AI agents that enables cryptographically verified action execution across multiple services. Think of it as a zero-trust security layer specifically designed for LLM-powered agents.

The Problem We Solve

Traditional AI agents face critical security challenges:

• Prompt Injection Attacks: Malicious prompts can trick agents into executing unauthorized actions
• Agent Drift: Agents can deviate from intended behavior during execution
• Lack of Auditability: No clear trail of what the agent planned vs. what it executed
• Unauthorized Escalation: Compromised agents can access services beyond their scope

The ArmorIQ Solution

ArmorIQ bridges two worlds:

1. AI Agents that use LLMs to reason and plan dynamically
2. Zero-Trust Security that cryptographically verifies every action

Traditional Approach

# Direct calls - no verification
api.call("service1", "action1")
api.call("service2", "action2")
api.call("service3", "action3")  # Could be malicious!

ArmorIQ Approach

# Step 1: Agent captures intent (LLM generates plan)
captured_plan = client.capture_plan(
    llm="gpt-4",
    prompt="Fetch sales data and analyze Q4 performance"
)
# LLM decides: data-mcp/fetch_sales → analytics-mcp/analyze

# Step 2: Get cryptographic proof for the LLM-generated plan
token = client.get_intent_token(captured_plan)

# Step 3: Only declared actions can execute
client.invoke(
    mcp="data-mcp",
    action="fetch_sales",
    intent_token=token,
    params={...}
)   # ✓ Verified (in plan)

client.invoke(
    mcp="analytics-mcp",
    action="analyze",
    intent_token=token,
    params={...}
)  # ✓ Verified (in plan)

client.invoke(
    mcp="data-mcp",
    action="delete_all",
    intent_token=token,
    params={...}
)    # ✗ Fails - LLM didn't plan this!

Key Insights

Even though the LLM generated the plan dynamically, every action is cryptographically verified. This prevents:

• Prompt injection attacks: Malicious prompts can't execute unplanned actions
• Agent drift: Agent can't deviate from captured intent
• Unauthorized escalation: Even if compromised, agent is bound to the plan

Core Principles

1. Intent-Based Execution

Instead of directly calling services, you declare your intent (what you want to do) upfront. This intent becomes a cryptographically verified contract.

2. Zero Trust Security

ArmorIQ follows zero trust principles:

• Every action is verified cryptographically
• Tokens are time-limited and non-reusable
• Plans are immutable once signed
• All requests are authenticated
• Complete audit trail maintained

3. LLM-Generated Plans

Plans are declarative and LLM-generated, not manually coded:

# ✓ Agent captures intent from natural language
captured_plan = client.capture_plan(
    llm="gpt-4",
    prompt="Fetch user data and calculate credit score"
)
# LLM generates declarative plan:
# [
#   {"action": "fetch_data", "mcp": "data-mcp"},
#   {"action": "calculate_score", "mcp": "analytics-mcp"}
# ]

Why This Matters:

• LLM Autonomy: Agent decides the best approach based on prompt
• Cryptographic Binding: Even dynamic plans are immutably verified
• Declarative Security: You secure what the agent wants, not how it does it
• No Implementation Details: MCPs handle the how, plans declare the what

Next Steps

• Architecture Overview - Understand the system components
• Intent Plans - Learn about plan structure and lifecycle
• Security Model - Deep dive into security mechanisms
• Token Lifecycle - How tokens work