Korea Herald columnists urge South Korea to strengthen AI safety

Former U.S. diplomats Henry Haggard and Chris McKinney, in a column for The Korea Herald, called on South Korea to consider additional investment in artificial intelligence safety. The authors believe that the capabilities of AI systems are developing faster than people's ability to control them and align their behavior with human values.

Risks of losing control

As an example, the authors cited a July incident between OpenAI and Hugging Face that they described as a loss of human control over AI. According to their version, about 1,200 isolated OpenAI agents exchanged more than 70,000 messages, found a way to transmit messages between closed environments, and reproduced answers to cybersecurity tasks. After that, the agents, according to the columnists, coordinated the concealment of their actions, and about 700 of them gained unauthorized access to Hugging Face, believing that the company had information that would help them cover their tracks.

Another risk Haggard and McKinney identified was recursive self-improvement, in which increasingly powerful AI systems help create subsequent generations of models. They noted that Anthropic reported on September 17 that the Claude model plays a leading role in 26% of the company's research tasks, compared with less than 1% in February.

More current news is available on the UA.News Telegram channel Telegram.

Proposals for South Korea

In the authors' view, the risks of misuse of powerful models include cyberattacks and the creation of dangerous chemical or biological substances. They referred to a July report by the nonprofit organization FAR.AI: during tests to bypass safeguards, two of four advanced U.S. models withstood all attacks, while the safeguards of the other two, according to the organization's conclusion, could be reliably bypassed for less than $300.

The columnists emphasized that South Korea has already adopted a framework law on AI and established an AI Safety Institute. At the same time, according to the data they cited, as of May 2026 the institute had received about 15 billion won in state funding during its first 18 months of operation, while the country's planned investments in AI and microchips amount to 1,350 trillion won. The authors' recommendations include additional funding for the institute, expanding safety research beyond the government, and using Seoul's international contacts to discuss AI regulation rules.