ShadowGlass

ShadowGlass is an innovative Chrome side-panel extension designed specifically for security and IT teams seeking to gain insight into the AI providers and models their organization interacts with. Unique to ShadowGlass is its commitment to privacy: all processing occurs locally within the browser, ensuring that no log data leaves your machine. This capability allows users to analyze and inventory their firewall or proxy logs without the risk of exposing any sensitive information.

Key Features

• Local Analysis: ShadowGlass reads log files directly from your browser and performs all analysis without making any network requests, guaranteeing data privacy.

• Comprehensive Insights: By simply dropping your .xlsx, .csv, or .jsonl firewall or proxy log exports into the side panel, users can utilize a reliable deterministic regex matching system to reveal hidden AI providers and usage patterns.

• Detailed Reporting: The extension generates a clear inventory of AI interactions, breaking down calls to high-level platforms like AWS and Azure down to specific underlying models. Anything that deviates from your pre-approved list is flagged for review as shadow AI.

• On-Device AI Assistance: When the built-in Gemini Nano from Chrome is available, it provides plain-English narratives and answers to questions about the findings. This ensures users can easily understand complex data without compromising security.

• Export Options: Users can export the generated inventory and accompanying narratives in various formats, including Markdown, self-contained HTML reports, or CSV files.

Privacy-First Design

ShadowGlass operates with minimal permissions, requesting only sidePanel and storage. There are no host permissions, no content scripts, and no outbound requests allowed. This design ensures that the extension does not invade user privacy or compromise log security.

Limitations

While ShadowGlass is powerful, it is important to acknowledge its limitations:

• Detection is based solely on URL patterns, which means if relevant model names are not included in the URL, they may be missed.
• Self-hosted tools without distinctive hosts are matched on a best-effort basis and are considered low-confidence candidates.
• Features dependent on the Gemini Nano availability may not be accessible on all devices.

Community Contributions

The effectiveness of ShadowGlass relies on community involvement. Regular updates to the endpoint pattern registry help keep pace with the rapid evolution of AI providers. Users are encouraged to contribute new patterns through the provided guidelines.

Getting Started

For a hands-on experience, this repository includes sample log files that allow users to see ShadowGlass in action without requiring their own data.

In summary, ShadowGlass serves as a crucial tool for organizations committed to monitoring and managing their AI interactions securely and privately.