The global discourse on the potential threats posed by frontier artificial intelligence (AI) has highlighted significant deficiencies in South Korea's AI safety response capabilities, including personnel, budget, and essential infrastructure.
On September 28, Rep. Lee Jeong-heon of the Democratic Party revealed data obtained from the Ministry of Science and ICT and the AI Safety Research Institute (AISI), indicating that the 'AI Safety Evaluation Office' within AISI has only six personnel dedicated to evaluating and responding to AI threats. Out of a total of seven staff members, only six are involved in technical evaluations, excluding one responsible for legislation and norms.
In comparison to other major countries, South Korea's figures are notably lacking. The UK employs approximately 200 personnel, while the US plans to increase its workforce from 34 to 80. Japan is also expanding its current staff of 31 to 60 this year, with a final goal of 200.
The budget disparities are equally stark. AISI, currently a subordinate department of the Electronics and Telecommunications Research Institute (ETRI), has an allocated budget of 8.9 billion won (approximately $7.5 million) for this year.
In contrast, the UK has allocated about 121.6 billion won (approximately $102 million), while the US has set aside around 23.2 billion won (approximately $19 million). Japan has significantly increased its budget from 3.4 billion won (approximately $2.8 million) to over 80 billion won (approximately $66 million) through supplementary budgets.
South Korea's infrastructure is also lacking. The UK’s safety research institute has access to a national 'compute' resource worth 1.5 billion pounds (approximately 2.7 trillion won) through government prioritization.
In contrast, South Korea has only five graphics processing units (GPUs), with two of them rented from private cloud servers.
Currently, AISI is structured as a department under ETRI, making it difficult to recruit top-tier AI security experts. There are also concerns about its inability to exercise independent regulatory and supervisory authority.
Moreover, under the current AI Basic Act, the maximum fine for violating safety correction orders is only 30 million won (approximately $25,000), leaving no effective means to respond to threats from global tech giants.
Rep. Lee stated, "To establish an AI security system that can respond to visible threats, amending the AI Basic Act is necessary. We must grant AISI legal independence and ensure it has its own budget and personnel authority to enhance its practical response capabilities." He added, "Relying solely on ineffective fine regulations against global tech companies does not guarantee public safety. We need to realize punitive measures that can effectively control and correct issues when threats arise and build a comprehensive national AI security governance framework."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.