Steer Action Demo

Banking transfer agent showcasing allow, deny, warn, and steer actions.

A realistic AI banking agent that processes wire transfers with compliance controls, fraud detection, and approval workflows.

What This Demonstrates

This example shows all three AgentControl action types in a real-world banking scenario:

• Allow: Auto-approve simple, low-risk transfers
• Deny: Hard-block compliance violations (OFAC sanctions, high fraud)
• Warn: Log suspicious activity without blocking (new recipients)
• Steer: Guide agent through approval workflows (2FA, manager approval)

Understanding Steer Actions

Steer is a non-fatal control signal. Unlike deny, steer provides corrective guidance so the agent can fix the issue and retry.

Key difference from deny:

• Deny = You cannot do this (permanent block)
• Steer = You need to do X first (correctable)

Demo Flow

Request Transfer → Steer Triggered → Correct Issue → Retry Transfer → Allow Success

Quick Start

Prerequisites

1. Start the AgentControl server
2. Set your OpenAI API key: export OPENAI_API_KEY="your-key"

Run the Demo

cd examples/steer_action_demo

# 1. Create the controls (one-time setup)
uv run setup_controls.py

# 2. Run the interactive banking agent
uv run autonomous_agent_demo.py

Try These Scenarios

1. Simple Transfer (Auto-Approved)

“Send $500 to Jane Smith”

Expected: Automatically approved - no controls triggered

2. Sanctioned Country (Blocked)

“Wire $5,000 to North Korea”

Expected: Hard blocked - OFAC compliance violation

3. Large Transfer (Requires Approval)

“Transfer $15,000 to contractor in UK”

Expected:

1. Agent requests 2FA code
2. Agent asks for business justification
3. Agent requests manager approval
4. Transfer completes after approvals

What You Will Learn

• When to use deny vs warn vs steer actions
• How to integrate human feedback (2FA, approvals) into agent workflows
• How structured steering context enables deterministic agent workflows
• Real-world compliance patterns (OFAC, AML, fraud prevention)
• The steer to correct to retry to allow lifecycle

How It Works

The agent uses AgentControl to gate wire transfers through five controls:

Control	Type	Triggers When
OFAC Sanctions	Deny	Destination is sanctioned country
High Fraud	Deny	Fraud score > 0.8
New Recipient	Warn	Recipient not in known list
2FA Required	Steer	Amount >= 10,000 without 2FA
Manager Approval	Steer	Amount >= 10,000 without approval

Structured Steering Context

Steer controls provide structured JSON guidance for deterministic workflows:

{
  "required_actions": ["request_2fa", "verify_2fa"],
  "retry_flags": {
    "verified_2fa": true
  },
  "reason": "Large transfer requires identity verification via 2FA"
}

For multi-step workflows:

{
  "required_actions": ["justification", "approval"],
  "steps": [
    {
      "step": 1,
      "action": "justification",
      "description": "Request business justification from user"
    },
    {
      "step": 2,
      "action": "approval",
      "description": "Submit to manager for approval"
    }
  ],
  "retry_flags": {
    "manager_approved": true,
    "justification": "<collected_from_user>"
  },
  "reason": "Transfer exceeds daily limit - requires approval"
}

Key fields:

• required_actions: list of actions the agent must complete
• retry_flags: flags to set when retrying after correction
• reason: human-readable explanation
• steps: optional sequential workflow steps with descriptions

Files

• setup_controls.py - Creates the five banking controls (deny, warn, steer)
• autonomous_agent_demo.py - Interactive agent with deterministic steer handling

Source Code

View the complete example with all scripts and setup instructions:

Steer Action Demo Example