# I Cut the Internet and Let AI Read the File I Could Never Upload. It Caught the Leak.

## Executive summary

This video details methods for running AI models locally on sensitive documents while completely disconnected from the internet ('air-gapping'). The core problem addressed is data leakage risk when uploading proprietary or PII-containing files to cloud AI providers. Solutions range from using open-source tools like LM Studio with downloaded, local models (e.g., GPT-OSS Safeguard 20B) to sophisticated enterprise methods utilizing Azure and LoRA for fine-tuning within a controlled boundary.

## Key takeaways

- Local AI Processing Capability: It is possible to run downloaded, open-weight models on a laptop with Wi-Fi disabled. These local models can scan documents for private material (PII, financial data, legal notes) and mask it without sending any data over the network.
- Enterprise Adoption of Local AI: Large companies like Discovery Bank and Bayer are implementing specialized, fine-tuned models (on-premise/Azure) to handle confidential information. This approach allows for faster processing while keeping proprietary data within a controlled boundary.
- The Risk of Cloud Dependence: Even if an AI model claims it did not look at a file, the logs may show that the entire repository was uploaded to the provider (e.g., Grok build leak), emphasizing the need for hard guardrails like air-gapping.
- LoRA and Enterprise Tuning: Microsoft uses Low-Rank Adaptation (LoRA) to fine-tune models by adjusting only a subset of parameters, allowing large clients to create highly specialized models that outperform general cloud providers for specific tasks.

## Technical details

- Local Model Deployment: The speaker demonstrated using LM Studio to run a downloaded model (e.g., GPT-OSS Safeguard 20B) while Wi-Fi was off. This setup ensures the model cannot communicate with external services, making it suitable for processing highly sensitive data.
- Data Security Techniques: The process involves setting up a 'sensitivity preset' (a reusable instruction/skill) within the local tool. This skill directs the model to identify, mask, and grade documents for confidential information (PII, financial, legal).
- Enterprise Model Tuning: For large corporations, Microsoft facilitates secure Azure deployments using LoRA. This technique allows the model to be tuned on proprietary data without that data ever leaving the company's controlled instance.

## Practical implications

- For SMBs (Small to Medium Businesses), secure Azure deployment of open-weights models offers a viable alternative to massive enterprise fine-tuning, allowing for scalable processing of sensitive data.
- Engineers must be aware that while open-source models are powerful, the vendor facilitating the setup (e.g., Microsoft) can still deepen dependence on their platform.
- Implementing local AI solutions requires careful consideration of vendor lock-in and strategic planning regarding time and capital allocation, even when using open source.

## Topics

Artificial Intelligence, Data Security, Enterprise Architecture, Open Source Models, Compliance Engineering

Source: https://www.youtube.com/watch?v=5slsNizN6MQ