CyCraft XecGuard

CyCraft XecGuard is an AI guardrail safety module designed to protect Large Language Models (LLMs) from malicious attacks. The product addresses security risks identified in OWASP 2025, specifically focusing on prompt injection, prompt extraction, and jailbreak attacks. XecGuard is built on a LoRA (Low-Rank Adaptation) adapter architecture that can be deployed on existing AI applications without requiring extensive modifications. The module enhances instruction-following accuracy in LLMs, enabling them to detect and block malicious contexts that attempt to violate system prompts. The product is designed to work with mainstream open-source LLMs including Llama 3B, Qwen3 4B, Gemma3 4B, and DeepSeek. According to testing data, XecGuard improves overall security defense scores by an average of 19.4%, with defensive capabilities reaching up to 33.9% enhancement against certain attack types. XecGuard maintains compatibility with common AI chatbot interfaces, allowing for deployment without significant impact on the model's original capabilities. The product is positioned for enterprise use across government, financial services, semiconductor, medical, and retail sectors. The solution includes LLM Red Teaming assessment capabilities to evaluate security resilience against various attack scenarios. XecGuard operates as a next-generation AI firewall layer that sits between user inputs and the LLM to filter malicious content before it reaches the model.