South Korea's Ministry of Science and ICT is actively sounding out participation in Anthropic's cybersecurity consultative body, Project Glasswing. Variables such as the White House's opposition to expanding the project and the lack of a memorandum of understanding (MOU) between Anthropic and the government leave the outlook uncertain.

On May 11, the ministry held an AI and cybersecurity cooperation meeting with Anthropic officials including Michael Sellitto (마이클 셀리토), Anthropic's head of global policy, together with the foreign ministry, the National Intelligence Service, the Financial Services Commission, the AI Safety Institute (AISI) and the Korea Internet & Security Agency (KISA), as well as the Financial Security Institute. Ryu Je-myung (류제명), the ministry's second vice minister, Kim Myung-joo (김명주), head of AISI, and Oh Jin-young (오진영), a KISA director, attended as South Korean representatives.

At the meeting, the ministry was reported to have conveyed to Anthropic its intention for broad cooperation including Glasswing. It proposed a joint response in cybersecurity with South Korean companies and institutions and asked for information sharing to help prepare in advance for vulnerability disclosures. Participants also discussed measures to implement AI safety and trust policies including the Basic AI Act, and cooperation between AISI and Anthropic.

A ministry official said, "As Korea is a country that has established comprehensive AI legislation following the EU, global companies are showing strong interest in how it will be implemented," adding, "We expressed our intention that Korea can contribute in this process."

Glasswing is a cybersecurity consultative body that Anthropic launched on April 7. It is operated so that participating companies and institutions jointly detect vulnerabilities and strengthen defensive capabilities based on Anthropic's latest AI model, 'Claude Mythos Preview.' About 50 companies and institutions are participating, including 12 launch partners such as AWS, Microsoft, Google, Apple and Nvidia. Mythos has been disclosed on a limited basis only to Glasswing participants, and among institutions outside the United States, the UK's AI Safety Institute (AISI) is the only one granted access. Britain signed an AI safety cooperation MOU with the United States in 2024.

The ministry is also sounding out participation in Glasswing through its affiliated AI Safety Institute. Vice Prime Minister Bae Kyung-hoon (배경훈), who also serves as science and ICT minister, said on his social media on April 28, "The government will also respond quickly in line with the changing cybersecurity paradigm," and "We will also closely review participation in global cooperation such as Anthropic's Glasswing project, centered on the AI Safety Institute."

But the possibility of participation remains unclear. According to foreign media including the Wall Street Journal, the White House is opposing an expansion of Glasswing due to Mythos' ripple effects. The Trump administration has previously clashed with Anthropic by demanding it fully open up the scope of Claude's military use. Anthropic also said it is continuing discussions on Glasswing with bodies including the U.S. Cybersecurity and Infrastructure Security Agency (CISA) and the Center for AI Standards and Innovation (CAISI). The industry has said that under such circumstances, it would be difficult for Anthropic to independently decide participating countries or companies.

◆ MOU with Anthropic cited as prerequisite..."To be fleshed out after Seoul office is established"

If Anthropic further expands Glasswing participants, an MOU could also be cited as a prerequisite. Anthropic currently has MOUs with the United States, Britain, Japan and Australia, among others. It signed agreements with Japan's AI Safety Institute in October last year and with the Australian government in April.

Kim Myung-joo, head of the Korea AI Safety Institute, said, "Even countries that have signed MOUs with Anthropic, such as Japan and Canada, have not been able to participate in Glasswing, and Korea is in a situation where it has to clear even the MOU hurdle first."

The ministry is speeding up efforts to sign an MOU with Anthropic that it has pursued since before the Mythos issue. After Bae met Anthropic CEO Dario Amodei at the 'AI Impact Summit' held in New Delhi, India, in February, it pushed for an MOU. The plan was to formalise cooperation after appointing the head of Anthropic's Korea office. But the establishment of Anthropic's Seoul office, which had been announced for early this year, has been delayed, and the MOU has also been postponed.

A ministry official said, "Anthropic's Korea office is expected to be established during June," adding, "Once the office is established, the MOU will also show concrete progress."

Separate from joining Glasswing, the ministry is also moving quickly to respond to AI cybersecurity threats. It plans to share vulnerabilities found through simulated attacks based on Claude Opus 4.7 with domestic experts and to release related measures as early as late this month. Some South Korean companies are also taking part in OpenAI's cybersecurity consultative body, 'Trusted Access for Cyber (TAC),' which it operates with 'GPT-5.4 Cyber.'