Security AI Goes Local: Fine-Tuning a Reasoning Model for Data-Sensitive Work

Security professionals face a critical constraint: cloud-based LLMs can't be trusted with sensitive client data or internal logs. This fine-tuned 1.5B reasoning model eliminates that trade-off, delivering genuine chain-of-thought reasoning entirely offline on modest hardware, while remaining compact enough (1.2GB) to fit on a USB drive. It specifically targets modern threats like MCP tool poisoning and AI-native jailbreaks that emerged in 2025-2026.