Yonhap reports that South Korea's Ministry of Science and ICT held a working-level workshop with OpenAI on May 18, 2026, where Sasha Baker, head of national security policy for ChatGPT, introduced Trusted Access for Cyber (TAC) and related cybersecurity functions of the company's latest models. Aju Press reports that on May 27 Vice Minister Ryu Je-myung met OpenAI Chief Strategy Officer Jason Kwon to formalize Seoul's participation in TAC, and that the Korea Internet & Security Agency (KISA) will serve as the implementing body. Aju Press also reports Korea and Japan are among the first two Asian countries to join the program. Aju Press published a direct quote from Vice Minister Ryu: "Through this cooperation with OpenAI, Korea has laid the groundwork to get ahead of AI-driven cyber threats." Editorial analysis: Industry observers should view this as part of a broader pattern where governments seek vetted, controlled access to frontier models for defensive cyber research and vulnerability assessment.
What happened
Yonhap reports that South Korea's Ministry of Science and ICT and related government bodies held a working-level workshop with OpenAI on May 18, 2026, focusing on AI-related cybersecurity issues. Yonhap reports Sasha Baker, who leads national security policy for ChatGPT, introduced cybersecurity-related functions in OpenAI's latest models and presented the company's Trusted Access for Cyber (TAC) program. Aju Press reports that on May 27 Vice Minister Ryu Je-myung met OpenAI Chief Strategy Officer Jason Kwon in Seoul to formalize Korea's participation in TAC, and that the Korea Internet & Security Agency (KISA) will act as the implementing body. Aju Press reports South Korea and Japan are among the first two Asian countries to join the program. Aju Press carries a direct quote from Vice Minister Ryu: "Through this cooperation with OpenAI, Korea has laid the groundwork to get ahead of AI-driven cyber threats." Yonhap reports attendees included officials from the Ministry of Foreign Affairs, the National Intelligence Service and the Financial Services Commission.
Editorial analysis - technical context
Public reporting frames TAC as a vetted-access program that provides governments and critical-sector institutions controlled entry to advanced models for defensive cyber work. Industry-pattern observations: governments and security agencies typically request such access to test how advanced models can both augment defensive tooling and be misused to automate or scale attacks. For practitioners, vetted access usually enables more realistic redteam exercises, controlled data sharing for model evaluation, and structured feedback loops with model providers without exposing unrestricted API access.
Context and significance
Industry context: Aju Press frames Seoul's move alongside competing, coalition-style efforts such as Project Glasswing, led by Anthropic and described by Aju Press as pairing the unreleased Claude Mythos Preview with launch partners and organizations to hunt vulnerabilities. The reported bilateral engagement with OpenAI fits a recent pattern where national authorities pursue provider-specific arrangements to gain timely technical insight into frontier-model capabilities and failure modes rather than relying solely on multilateral coalitions. For national cybersecurity teams and private-sector defenders, earlier visibility into model capabilities can shorten the window for discovering novel attack vectors and for validating defensive mitigations.
What to watch
Observers should track:
- •whether KISA publishes technical guidance or joint evaluation reports stemming from the OpenAI engagement
- •any formal terms or timelines that Aju Press or Korean agencies release for TAC access and which classes of institutions are approved
- •whether Seoul pursues parallel participation in initiatives like Project Glasswing (Aju Press reports Korea is exploring such options). Industry-pattern observations: transparency around evaluation methods and capability limits will materially affect how security teams incorporate model-assisted detection, threat hunting, and vulnerability scanning into existing tooling
Scoring Rationale #
Notable for security practitioners: government-level vetted access to frontier models affects national defensive capabilities and coordination between model providers and public agencies. The story alters how security teams anticipate threat modeling and vulnerability discovery.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.