The government is responding to the Mythos Shock, which suggests that artificial intelligence (AI) will upend existing security paradigms. They are particularly exploring ways to gain access to Anthropic’s AI model, Claude Mythos. However, some critics argue that Anthropic is exaggerating AI’s hacking threats, sparking a controversy over fear marketing.
According to government officials on Wednesday, the Ministry of Science and Information and Communications Technology (ICT) is considering participating in Anthropic’s Project Glasswing to gain access to Mythos.
The South Korean government is currently in talks with Anthropic to secure access to Mythos. Deputy Minister Ryu Je-myung stated during a National Assembly committee meeting the previous day that they are exploring participation in the Glasswing project.
Deputy Prime Minister and Minister of Science and ICT Baek Beom-kyu also confirmed on May 27 that they are actively engaging with Anthropic.
Earlier this month, Anthropic unveiled Mythos, a general-purpose AI model with autonomous security intelligence. This model can demonstrate security capabilities based on agentic coding and reasoning abilities without specialized cybersecurity training. It can identify vulnerabilities and even design infiltration paths by reasoning at human expert levels.
Notably, Anthropic has not publicly released Mythos. Instead, it’s providing access to just 52 companies and organizations as part of Project Glasswing, which aims to address security challenges in the AI era. Anthropic claims Mythos proved its capabilities by finding a 27-year-old bug in OpenBSD, an operating system known for high security, with just a single command to find weaknesses and infiltrate.
As a result, the global financial sector is on high alert. There are growing concerns that if AI security threats materialize, they could paralyze financial infrastructure.
The South Korean government has also begun its response. On April 14, the Financial Supervisory Service urgently convened information security personnel from major financial institutions to review response measures. The National Security Office at the Blue House has ordered urgent responses from relevant ministries. Consequently, the Ministry of Science and ICT has held four emergency meetings to address current issues.
Deputy Prime Minister Baek announced through his social media account that they have confirmed, within disclosable limits, that the latest model’s cyber capabilities allow for significant vulnerability exploration and attack scenario creation with just prompting, without advanced coding.
Additionally, they plan to prioritize responses to large-scale vulnerabilities, enhance anomaly monitoring, and strengthen security checks for critical infrastructure and public systems. They also indicated consideration of participating in global collaborations like Anthropic’s Glasswing project, drawing inspiration from the UK’s AI Security Research Institute (AISI), which has gained access to Mythos.
In the U.S., Controversy Over Mythos’s Fear Marketing Continues… Risks May be Exaggerated
As concerns about the Mythos Shock rapidly spread, some in the U.S. have raised issues with Mythos’s fear marketing. Initial users claim that the risks associated with Mythos have been overstated and that it does not surpass highly skilled security professionals’ capabilities.
According to major media outlets like The New York Times, early users who experienced the preview version of Mythos have stated that it is not yet capable of replacing human security researchers.
Critics argue that Anthropic’s secretive approach to the project has led to exaggeration of the technology’s impact, fueling fear through an AI marketing strategy.
Yann LeCun, a professor at New York University and a prominent AI figure, remarked that the drama surrounding Mythos is nothing more than fiction born of self-deception.
Sam Altman, Chief Executive Officer (CEO) of OpenAI, criticized Anthropic’s actions, stating that there have long been people who wanted to keep AI in the hands of a few, and accused them of employing a fear-based marketing strategy by creating a bomb and selling shelter only to their chosen customers.
