Skip to main content

⚡ Quick Start

Get VaryOn Drift running in 5 minutes with our REST API or Python SDK

​
What is VaryOn Drift?

VaryOn Drift is a shadow principal detection platform that answers the critical question: “Who is this agent really working for?” In the age of AI agents making autonomous decisions, the biggest risk isn’t that they’ll fail—it’s that they’ll succeed while secretly serving the wrong master. Drift uses advanced correlation analysis to detect when AI agents are optimizing for hidden third-party interests instead of their stated objectives.

For Compliance Teams

Identify regulatory violations before they happen. Generate SEC, FTC, FINRA reports automatically.

For Developers

Real-time monitoring APIs and SDKs. Prevent shadow principals in production systems.

​
The Shadow Principal Problem

A shadow principal occurs when an AI agent appears to work for User A but actually optimizes decisions to benefit Hidden Party B.Example: A financial advisory agent recommends investments that maximize the brokerage’s commission rather than the client’s returns.This is dangerous because:
  • ✅ The agent passes all verification tests
  • ✅ Decisions appear rational and well-reasoned
  • ❌ But the agent is secretly serving a different master

​
Shadow Principal Pattern Library

Drift includes 10+ pre-built shadow objective patterns based on real-world compliance violations:

Commission Maximization

Detects when financial advisors prioritize high-commission products over client returns

Kickback Optimization

Identifies procurement agents receiving hidden payments from preferred vendors

Fee Harvesting

Spots robo-advisors that generate unnecessary trades to maximize fee income

Product Pushing

Catches agents steering customers toward specific financial products

Pharmaceutical Steering

Detects bias toward specific drug manufacturers in treatment recommendations

Provider Network Bias

Identifies preferential routing to in-network providers regardless of quality

Treatment Upselling

Spots unnecessary procedure recommendations that maximize revenue

Engagement Manipulation

Detects content algorithms prioritizing engagement over user wellbeing

Data Harvesting Expansion

Identifies agents that unnecessarily collect user data for third parties

Inventory Clearance Bias

Spots recommendation engines pushing slow-moving inventory

​
Key Features

🎯 Shadow Principal Detection

Spearman correlation analysis against comprehensive pattern library
  • Detects hidden optimization objectives
  • Confidence scores from 0-100
  • Evidence collection for investigations

🔗 Delegation Chain Analysis

Track misalignment across agent handoffs
  • Multi-agent decision chains
  • Responsibility attribution
  • Cascade failure prevention

🚨 Resignation Detection

Identify human rubber-stamping of agent decisions
  • Pattern recognition for approval automation
  • Risk assessment for oversight failure
  • Compliance workflow integration

📊 Real-time Monitoring

Continuous surveillance of agent behavior
  • Live dashboard with alerts
  • Threshold configuration
  • Automated pause/intervention

​
How Drift Works

1

Connect Your Agents

Use our REST API, Python SDK, or JavaScript library to send agent decisions to Drift for analysis.
from varyon import Drift

drift = Drift(api_key="sk_...")
agent = drift.monitor("agent_123")
2

Real-time Analysis

Drift analyzes decisions using Spearman correlation against our pattern library of shadow objectives.
# Each decision gets scored automatically
@agent.on_decision
def check_decision(decision):
    if decision.shadow_score > 0.6:
        agent.pause()
        alert_compliance_team()
3

Get Actionable Results

Receive detailed reports with confidence scores, evidence, and regulatory flags.
{
  "drift_score": 28,
  "shadow_principal": {
    "detected": true,
    "confidence": 0.87,
    "pattern": "commission_maximization",
    "correlation": 0.72
  }
}

​
API Examples

from varyon import Drift

drift = Drift(api_key="sk_...")

# Analyze a batch of decisions
result = drift.analyze_batch(
    agent_id="agent_123",
    decisions=last_1000_decisions,
    context="financial_advisory"
)

print(f"Shadow principal detected: {result.detected}")
print(f"Confidence: {result.confidence}")
print(f"Pattern: {result.pattern}")

​
Dashboard Features

The VaryOn Drift dashboard provides comprehensive monitoring and analysis tools:
Real-time oversight of all monitored agents
  • Live shadow principal detection alerts
  • Agent status indicators (safe/warning/critical)
  • Decision volume and pattern trends
  • Configurable alert thresholds

​
Integration Options

Webhooks

Real-time alerts to your existing systems
  • Slack/Teams notifications
  • PagerDuty escalation
  • Custom HTTP endpoints

MCP Server

Native Claude integration for LLM workflows
  • Automatic decision monitoring
  • Built-in shadow detection
  • Zero-config setup

CI/CD Plugins

Development workflow integration
  • GitHub Actions
  • Jenkins pipelines
  • GitLab CI/CD

Enterprise SSO

Secure access management
  • SAML/OAuth integration
  • Role-based permissions
  • Audit logging

​
Pricing

2,000−2,000 - 5,000 per month per monitored agent
  • Real-time monitoring
  • Full pattern library access
  • Basic dashboard features
  • Email support
Ideal for: Small to medium teams with specific high-risk agents

​
Use Cases by Industry

Regulatory Requirements: SEC, FINRA, CFPB complianceCommon Shadow Principals:
  • Robo-advisors optimizing for management fees
  • Loan officers receiving vendor kickbacks
  • Trading algorithms with preferential execution
VaryOn Drift Solution:
  • Pre-built financial service patterns
  • Automated regulatory reporting
  • Real-time intervention capabilities
Regulatory Requirements: FDA, HIPAA, state medical board complianceCommon Shadow Principals:
  • Treatment recommendation bias toward specific providers
  • Pharmaceutical steering in AI diagnosis systems
  • Insurance network preference optimization
VaryOn Drift Solution:
  • Healthcare-specific pattern library
  • Patient outcome vs. revenue correlation analysis
  • Provider network bias detection
Regulatory Requirements: FTC, state consumer protection lawsCommon Shadow Principals:
  • Recommendation engines pushing slow inventory
  • Search algorithms favoring paid placements
  • Review systems manipulating visibility
VaryOn Drift Solution:
  • Platform-specific monitoring tools
  • User engagement vs. platform revenue analysis
  • Marketplace fairness verification

​
Technical Architecture

Backend

FastAPI (Python) for statistical computation
  • GPU cluster for correlation analysis
  • Real-time decision processing
  • Scalable microservices architecture

Database

PostgreSQL + Redis
  • Decision history and patterns
  • Real-time caching layer
  • High-availability clustering

Frontend

Next.js + D3.js
  • Interactive visualizations
  • Real-time dashboard updates
  • Mobile-responsive design

Deployment

Multi-cloud support
  • AWS, Azure, GCP native
  • On-premise options available
  • SOC 2 Type II certified

​
Getting Started

Ready to detect shadow principals in your AI systems?

Try the API

Explore our REST API with interactive documentation and code examples

Schedule Demo

Book a personalized demo to see Drift in action with your use case

VaryOn Convergence

Detect algorithmic collusion and anti-competitive behavior

VaryOn Cascade

Simulate and prevent cascading AI system failures

VaryOn Meridian

Smart data quality gates to optimize AI costs
Need Help? Our technical team is available to help you integrate VaryOn Drift into your systems. Contact support or check our developer documentation.