An unrestricted attack based on diffusion models that can achieve both good transferability and imperceptibility.
Client SDK to send LLM interactions to Vibranium Dome
LLM Security Platform Docs
RAG pipeline red-teamed against prompt injection: poisoned document corpus, 20 attack payloads across 6 categories, LangGraph guard (sanitizer + LLM classifier + output validator). Baseline 40% attack success → 0% defended, +0.43s latency cost, honest false-positive analysis.
State-of-the-art adversarial attacks and defences for FLEXible framework
This project focuses on the development and analysis of Poison Attack and Evasion Attack strategies.
Add a description, image, and links to the adverarial-attacks topic page so that developers can more easily learn about it.
To associate your repository with the adverarial-attacks topic, visit your repo's landing page and select "manage topics."