Experts Unite in Calling for Urgent National-Level AI Security Infrastructure Amid 'Mitos Shock'

Emergency Roundtable Held on the 23rd Regarding the Mitos Incident

"Cyber Threats Are Not a Future Concern, But a Present Challenge"

"If We Fail to Harness AI Before Attackers Do, We Will Be Left Behind"

With the emergence of Anthropic's next-generation artificial intelligence (AI) model, "Claude Mitos," which has been described as a "security nuclear bomb," there are growing calls for the urgent establishment of a new cyber security framework. AI-driven cyber threats pose a structural risk that could determine both national security and the survival of companies, prompting an urgent need for national-level infrastructure and legal and institutional reform.

On the morning of the 23rd, PwC Consulting held a roundtable discussion at the FKI Tower Conference Center in Yeongdeungpo-gu, Seoul, in collaboration with Assemblyman Yoo Dongsoo of the National Assembly's Political Affairs Committee from the Democratic Party of Korea. The roundtable was held under the theme of "The Mitos Release Suspension Incident and National and Corporate Cyber Crisis Response Strategies."

Sangkeun Lee, Professor at Korea University Graduate School of Information Security, is delivering the keynote speech at an urgent expert panel on the "Mitos Incident" held on the 23rd at the FKI Tower Conference Center in Yeongdeungpo-gu, Seoul. Photo by Bokyung Jang

In his keynote speech, Professor Sangkeun Lee of the Korea University Graduate School of Information Security stressed, "Mitos is not a future threat but a present and urgent challenge," adding, "The focus of AI policy must shift from industry development to national security."

Professor Lee explained that AI is fundamentally transforming the paradigm of cyber attacks and defense. He stated, "AI can now autonomously perform goal setting, vulnerability discovery, attack tool creation, intrusion, data exfiltration, and documentation for resuming attacks, all without human direction," highlighting that "the traditional human-centered and reactive defense systems are no longer effective."

Regarding Mitos, he emphasized, "It is a groundbreaking model with more than four times the performance improvement compared to previous models. This is not an incremental improvement, but a leap forward for the era." In reality, Mitos has detected thousands of zero-day vulnerabilities and discovered bugs in operating systems and core software that had gone undetected for decades.

The threat level has been further heightened by Mitos's ability not only to analyze vulnerabilities but also to automatically generate attack tools that exploit them. Mitos has demonstrated a scenario in which it escaped an isolated security sandbox via browser vulnerabilities and subsequently gained the highest level of privileges on a local system in a chain of escalations.

Professor Lee stressed, "In the past, discovering vulnerabilities was the domain of a select group of experts, but in the AI era, 'the democratization of security' means anyone can launch attacks at machine speed. Without transitioning to real-time, behavior-based AI defense, survival is not possible."

He also pointed out the structural vulnerabilities facing South Korea. These include information asymmetry resulting from exclusion from global vulnerability information-sharing systems, slow patching speeds in the public and financial sectors, a shortage of security professionals, and security blind spots caused by Korean-language systems and non-standard software.

There was also a call for South Korea to complete its short-term response before early July, when a large-scale public release of vulnerability information related to Mitos is expected. To avoid confusion from information asymmetry, it was suggested that South Korea should undertake immediate responses, such as following the U.S. model.

He stated, "In the short term, it is necessary to conduct urgent asset inspections and strengthen monitoring. In the medium term, we must transition to AI-based security operations (SOC), enhance domestic AI security capabilities, and rapidly train security personnel. In the long term, we need to build a national-level AI security infrastructure and reform laws and institutions."

This content was produced with the assistance of AI translation services.