SANDS Lab builds a public-facing platform to counter deepfakes and other harmful uses of generative AI

SANDS Lab builds a public-facing platform to counter deepfakes and other harmful uses of generative AI

• Wins KRW 10 billion national project to develop technology that curbs harmful uses of generative AI
• Public services planned to counter emerging AI-enabled crime, including deepfakes, phishing and synthetic content
• Developing future cybersecurity technologies and measures to combat digital crime

Photo 1. Technology development for detecting and responding to harmful uses of generative AI (Provided by SANDS Lab)

AI-based cyber threat intelligence (CTI) specialist SANDS Lab (KOSDAQ 411080, CEO Kim Ki-hong) announced on the 28th that it is developing a cybersecurity-specific RAG-based sLLM and a public-facing platform to curb harmful uses of generative AI.

Harmful uses of generative AI refer to negative or unintended misuse of the technology, including misinformation and fake news creation, copyright infringement, deepfakes and cybercrime.

The government is therefore working to improve frameworks for ethical use of generative AI and develop core technologies so that everyone can use AI safely.

The Ministry of Science and ICT and IITP commissioned a KRW 10 billion core information security technology development project to create a cybersecurity-specific RAG-based sLLM and a public-facing platform that mitigate harmful uses of AI.

After winning the project in April, SANDS Lab began joint R&D with LG Uplus, 42Maru, Logpresso and the Korea Advanced Institute of Science and Technology (KAIST).

The core goals are technologies to prevent harmful AI uses such as deepfakes, voice phishing and voice synthesis, and an on-premises specialized large language model (sLLM). The project aims to minimize harm and improve digital safety and reliability, while shortening decision-support time through faster information retrieval and analysis and delivering user-centered security services through conversational interfaces.

Unlike typical security companies, SANDS Lab owns petabyte-scale datasets, a prerequisite for AI model development. Collected and preprocessed data needed for domain-specific AI can be used immediately for training and deployed as an integrated detection and response system. Building domain datasets, sLLMs and conversational interfaces gives it a significant advantage in R&D for diverse cybersecurity-specific AI applications.

The public-facing platform, one part of the R&D project, applies detection and response technologies to harmful generative AI uses. It is being developed to combat digital crimes and limit damage from phishing, fake news and misinformation, mass malware production, voice synthesis, and deepfake images and videos—issues that have become prominent social concerns.

According to SANDS Lab’s AI technology team, the deepfake video and image detection technology being developed with LG Uplus uses deep learning algorithms, adversarial training to counter generative adversarial networks (GANs), and multimodal data analysis for greater precision. Model compression and optimization will reduce complexity while maintaining or improving accuracy to enable immediate detection in live video streams and on social media platforms.

The platform for detecting deepfakes, phishing emails, malware, voice synthesis and fake news brings together detection technologies and other AI capabilities, including retrieval-augmented generation (RAG) to explain results and decryption. A mobile app to detect and block smishing and voice phishing is also under development. These tools are expected to enable early response and protect the public from emerging digital crimes, improve awareness, spread security practices and build trust in Korean security technology.