SANDS Lab unveils GLX, an ultra-fast LLM guard, advancing generative AI security performance and speed

SANDS Lab unveils GLX, an ultra-fast LLM guard, advancing generative AI security performance and speed

• New Security for AI technology for generative AI security debuts at the IASF 2025 seminar
• Addresses the slow speed and high resource consumption of conventional LLM-based guards with a lightweight implementation that requires no GPU
• Real-time security filtering at streaming speeds on CPUs, blocking even multi-turn attacks

Photo 1. SANDS Lab CEO Kim Ki-hong unveils GLX, an ultra-fast LLM guard, at IASF 2025.

SANDS Lab (KOSDAQ 411080; CEO Kim Ki-hong) announced that it unveiled its independently developed ultra-fast LLM guard, GLX, at the Integrated Application Security Fair (IASF) 2025, hosted by MONITORAPP on the 10th.

Now in its 13th year, the annual event presents the latest technologies and shares global security trends. SANDS Lab introduced its LLM guard as a core Security for AI technology, proposing a new standard combining speed and efficiency.

Generative AI is spreading rapidly across industries, but it is also creating new security threats. Large language models (LLMs) are particularly vulnerable to malicious prompt injection, sensitive data exposure, model poisoning and other attacks, making them a major challenge for generative AI security.

To address these issues, SANDS Lab independently developed GLX (Guard for LLM X), a dedicated LLM security solution. Following the company's CTX and MNX brands, GLX blocks potential misuse through attack simulations against a range of Korean and international LLMs, helping businesses and institutions use generative AI with confidence.

Many LLM guards currently on the market rely on calling another LLM to detect threats. This approach has drawn criticism for long response delays and excessive GPU consumption, making deployment in real-world services difficult.

GLX, however, applies proprietary lightweight technology to enable real-time security filtering on CPUs without a GPU. Its low-latency, streaming-level processing helps organizations strengthen generative AI security while maintaining stable service quality.

According to SANDS Lab researchers, internal testing showed that GLX was more than three times faster than existing overseas LLM security products at its best, while perceived speed improved by more than 50 times compared with GPU-based environments.

The LLM guard unveiled by SANDS Lab is commercial technology ready for immediate use in enterprise environments, rather than a research prototype. It operates at stable speeds in on-premises environments without GPU servers, reducing costs, and delivers both security and performance in multi-user cloud environments. Its lightweight architecture can also be integrated easily into existing AI services, improving deployment convenience and maximizing the range of possible applications.

It is designed to detect and block threats throughout three stages: input (user queries), response (LLM output), and retrieval-augmented generation (RAG). This technical achievement also aligns with the government's N2SF (National Network Security Framework) policy, enabling proactive responses to emerging attack techniques such as prompt injection, personal information leakage and sensitive data exposure.

Under its SANDS Lab Community vision of growing together with partners, announced earlier this year, SANDS Lab plans to launch GLX through joint development with MONITORAPP, Korea's leading web application firewall company. The company then plans to extend its guard technology beyond LLM security to generative AI domains such as images and audio, with the aim of leading the AI security market.

CEO Kim Ki-hong said, “To address LLM threats, a core challenge in Security for AI, SANDS Lab has developed an ultra-fast LLM guard that fits real-world AI security response frameworks. GLX differentiates itself from other LLM security products by providing defenses validated through diverse LLM attack simulations while enabling real-time responses without dependence on GPUs.”

Kim added, “Having developed it together with MONITORAPP, Korea's leading web application firewall company, we believe it will become an AI-focused security service that redefines industry standards. We will continue to expand our security technology into diverse generative AI fields and lead AI security innovation in the market.”

Photo 2. Part of SANDS Lab's LLM guard presentation, showing evaluation metrics indicating substantially greater efficiency than existing products.

Photo 3. SANDS Lab CEO Kim Ki-hong presents generative AI security technology at IASF 2025.