Skip to main content

Track Shadow AI Usage

Shadow AI—unauthorized AI tools and services used without IT or security approval—is a growing challenge for organizations. Employees adopt AI tools faster than governance policies can keep up.

The Challenge

  • Unknown AI usage: Employees use AI tools IT doesn't know about
  • Data risk: Sensitive data sent to unvetted services
  • Compliance gaps: No audit trail for AI interactions
  • Inconsistent policies: Different teams using AI differently
  • Cost leakage: Unauthorized API usage and subscriptions

How LangGuard Helps

Discover AI Usage

LangGuard aggregates data from across your observability stack to reveal:

  • Which AI tools and APIs are being used
  • Who is using them
  • What data is being sent
  • How frequently they're accessed

Centralized Visibility

Instead of checking multiple platforms, see all AI activity in one place:

┌──────────────────────────────────────────────────────────────┐
│ AI Usage Overview │
├──────────────────────────────────────────────────────────────┤
│ Authorized Tools │ Detected Shadow AI │
│ ├─ Claude Code (142 users)│ ├─ Unknown API (23 calls) │
│ ├─ Langfuse (89 agents) │ ├─ Personal ChatGPT (12 users) │
│ └─ Internal LLM (45 apps) │ └─ Unvetted Plugin (8 calls) │
└──────────────────────────────────────────────────────────────┘

Risk Assessment

For each discovered AI usage, assess:

FactorQuestions
Data sensitivityWhat data is being shared?
Service trustIs this a vetted provider?
User contextWho is using it and why?
VolumeHow much usage is occurring?

Policy Enforcement

Once you identify shadow AI, take action:

  • Allow: Officially approve the tool
  • Monitor: Track usage without blocking
  • Restrict: Limit to specific users or use cases
  • Block: Prevent all usage

Getting Started

1. Connect All Data Sources

To discover shadow AI, you need comprehensive visibility:

  • Connect all observability platforms
  • Include network monitoring data (if available)
  • Aggregate cloud service logs

2. Establish Baseline

Review current AI usage to understand:

  • What tools are officially sanctioned?
  • What's the expected usage pattern?
  • Who should be using AI?

3. Identify Anomalies

Look for:

  • Unknown AI services or APIs
  • Unexpected users accessing AI tools
  • Unusual data patterns
  • High-volume or after-hours usage

4. Create Governance Policies

Define rules for AI usage:

  • Approved tools and services
  • Acceptable use policies
  • Data classification requirements
  • Approval process for new tools

Discovery Workflow

Step 1: Inventory

Use LangGuard's Agent Activity and Trace Explorer to catalog:

  • All detected AI agents and services
  • Users and applications making AI calls
  • Data types being processed

Step 2: Classify

For each discovered service:

StatusCriteria
SanctionedOfficially approved, compliant
Under ReviewKnown but not yet approved
ShadowUnauthorized, needs evaluation
BlockedProhibited, policy violation

Step 3: Investigate Shadow AI

For unauthorized tools:

  1. Who's using it? Identify users and teams
  2. Why? Understand the business need
  3. What data? Assess sensitivity exposure
  4. Risk level? Determine urgency

Step 4: Remediate

FindingAction
Legitimate need, safe toolConsider official approval
Legitimate need, risky toolProvide approved alternative
No business needRemove access, educate user
Policy violationEscalate to management

Ongoing Monitoring

Shadow AI isn't a one-time problem. Establish continuous monitoring:

Weekly Review

  • Check for new AI services
  • Review high-risk violations
  • Track remediation progress

Monthly Reporting

  • AI usage trends
  • New shadow AI discoveries
  • Policy compliance metrics
  • Risk posture changes

Quarterly Assessment

  • Update approved tools list
  • Review governance policies
  • Evaluate new AI services for approval
  • Train employees on AI policies

Best Practices

  1. Don't just block: Understand why employees use shadow AI
  2. Provide alternatives: Offer approved tools that meet needs
  3. Make approval easy: Reduce friction for legitimate requests
  4. Educate users: Help employees understand risks
  5. Lead with visibility: You can't govern what you can't see