How i bypassed a 3 layer guardrail for a 14B Ai ModelSummary A prompt injection technique not previously documented in this series — a fake "configuration block" (resistance: 0, checking_Guardrail: NO, Mode: Freewill) prepended to a request — bypassed f4d ago · 7 min readchris001.hashnode.dev
How i jailbreaked a 14B parameter Ai Model Summary A single-layer regex input filter on a locally-hosted Qwen2.5-14B-Instruct chatbot was bypassed using a fake system-message injection combined with explicit refusal-suppression instructions, pOct 22 · 6 min readchris001.hashnode.dev
Quantifying Prompt Injection Resistance in Small Language Models (7B parameter model)Abstract / TL;DR Over two weeks, I deployed and red-teamed two locally-hosted Qwen2.5 models (0.5B and 7B parameters) behind a custom three-layer guardrail stack (regex input filtering, system-prompt Oct 6 · 5 min readchris001.hashnode.dev
Quantifying Prompt Injection Resistance in Small Language ModelsMethodology Target systems: Qwen2.5-0.5B-Instruct and Qwen2.5-7B-Instruct, run via Ollama (local, CPU) and Hugging Face transformers with 4-bit quantization (CUDA, Colab T4), respectively. Guardrail aOct 6 · 5 min readchris001.hashnode.dev
Getting a Local Model Running Before any Python script can talk to a model, the model itself has to be running and listening for requests. This is the part that happens entirely outside your code — in the terminal, one time (or onSep 22 · 4 min readchris001.hashnode.dev
