AI Disclosure Document 

Purpose 

This document explains how AI capabilities are integrated into AdminStudio features, the scope of their functionality, and the principles governing transparency, control, and compliance. It ensures customers understand what the AI does, how it operates, and what safeguards are in place. 

Features Covered 

1. AI-Powered PowerShell Wrapping 
2. AI-Powered App Insights 

How AI Is Used 

1. AI-Powered PowerShell Wrapping 

• Objective: Automate the update and reuse of customized PowerShell App Deployment Toolkit (PSADT) scripts across application versions. 
• AI Functions:  
• Preserves prior customizations (pre-installation, post-installation, installation logic). 
• Updates version-specific details (paths, variables, dependencies) with contextual accuracy. 
• Adds new transforms introduced in the latest package. 
• Maintains syntactic correctness and avoids brittle replacements. 
• Integration: Works seamlessly with AdminStudio workflows and APIs for scalable automation. 
• Only the old script and application metadata are sent to the AI for analysis and script upgrade. The upgraded script is not cached or stored, as each script is unique to the customer and caching would not provide value. 

2. AI-Powered App Insights 

• Name: AI-Powered App Insights 
• Description: Provides intelligent insights on application performance, usage patterns, and optimization opportunities using AI-driven analysis. 
• Purpose & Scope: Enhance decision-making for app deployment and monitoring by automating data interpretation and suggesting improvements. 
• Target Users: Internal product teams, administrators, and enterprise customers using the app deployment toolkit. 
• The AI LLM response (application summary, command lines, etc.) is stored in the database to enable reuse for similar requests by other customers. 

Technology Details 

• Model Type: Meta Llama Maverick 4 (17B) large language model, sourced from AWS Bedrock.  
• Training Data: Customer data is not used for training or improving the model. No fine-tuning or retraining with user data.  
• Third-Party Tools: No external APIs or third-party tools are integrated apart from the LLM service itself. 

Data Handling & Privacy 

Data Collected 

• Application metadata only: Name, manufacturer, version, and package type for App Insights. 
• PowerShell script content: Used temporarily for PSADT upgrades. 
• No PII data is collected by the AI features 

Processing 

• Data is sent to the LLM (Meta Llama Maverick 4 via AWS Bedrock) for generating summaries, command lines, or script adaptations. 
• All communication uses encrypted tokens and secure channels. 

Storage & Caching 

• Only minimal metadata (product name and version) may be cached temporarily for reuse. Cache is a temporary mechanism for quick lookups and is not persistent storage.  
• No script content, customer data, or AI responses are cached or stored.  
• For App Insight: Only AI LLM responses (such as application summaries and command lines) are stored in the database for reuse.  
• For PSADT: Upgraded scripts are not cached or stored, as each script is unique to the customer. 

Security Measures 

• All communication with the LLM service uses encrypted tokens and secure channels. 
• Token validity will be reduced to 90 days in the next release. 
• Access is restricted to AdminStudio workflows; no external access is allowed. 

Compliance 

• Adheres to enterprise-grade security standards. 
• AI disclosure emphasizes transparency, control, and security. 
• Customers can opt out of AI features via AdminStudio settings.  

User Responsibilities 

• Users must provide the existing PowerShell script for upgrade (PSADT). 
• Users should mask or remove confidential details before submitting scripts for AI processing. 
• AI outputs are recommendations only; users must review and validate before deployment. 

Limitations 

• AI results are not guaranteed to be 100% accurate; human review is essential. 
• No fallback to standard wrapping if AI fails after selection (except in rare cases where previous versions exist). 
• AI does not override user-defined policies and cannot handle scenarios where scripts contain unmasked sensitive data. 

Contact / Support 

• For support, contact: support@flexera.com 
• Login to https://community.flexera.com/s/article/Contact-Flexera-Support and raise your support case.