NVIDIA on the 27th officially launched the “Open Secure AI Alliance,” a global collaborative body for artificial intelligence (AI) security. Unlike existing security councils led by generative AI developers such as OpenAI and Anthropic, this alliance expands participation to include semiconductor, infrastructure and platform companies. In Korea, SK Telecom and Naver joined as founding partners.

Until now, discussions on AI safety and security have been led mainly by companies that develop AI models, such as large language models. Representative examples include “Project Glasswing,” a defensive security alliance organized by Anthropic in April around its high-performance model “Mythos,” and the “Frontier Model Forum,” established in 2023 by OpenAI, Anthropic, Google and Microsoft.

This time, NVIDIA, the market leader in AI computing semiconductors (GPUs), has taken the initiative in forming a cooperative body. The move is seen as an effort to expand security discourse, previously focused on AI models and software, to cover semiconductors, data centers and service operation stages.

The operating method also differs from that of existing councils. Unlike cases where a few companies have developed defensive technologies within closed systems, this alliance has decided to share security tools and technologies based on open source. The judgment is that, to respond to hackers who rapidly advance attack techniques using AI, it is more effective for multiple parties to jointly verify and reinforce technologies than for specific companies to monopolize them.

NVIDIA recently cited Hugging Face’s security incident as a case in point. At the time, closed AI tools imposed constraints on in-house analysis, but Hugging Face ran the open model “GLM 5.2,” developed by China’s Zhipu AI (Z.AI), on its own infrastructure, analyzed more than 17,000 activities and used the findings to block intrusions, according to the company.

More than 30 companies, including NVIDIA, Microsoft, IBM, Cisco, Palantir and CrowdStrike, are participating as founding partners. The Korean participants are SK Telecom and Naver. This aligns with NVIDIA’s recent moves to position Korea as a key AI hub in the Asia-Pacific region and to expand cooperation with both companies.

The participating companies plan to convert their proprietary security technologies into open formats and jointly identify and remedy vulnerabilities in AI systems. NVIDIA has already released tools that monitor and inspect the behavior of AI programs, while Hugging Face has disclosed technology for securely storing AI models.

SK Telecom has open-sourced its Korean language–specialized model “A.Dot X (A.X)” and has been pursuing a “sovereign AI” strategy under which data and infrastructure are managed directly within Korea. Naver is also working with NVIDIA to build a large-scale AI data center. An SK Telecom representative said, “We will work with global partners to help create a safe and trustworthy AI ecosystem.”

Popular News