What happened
The Korea Internet and Security Agency is developing new safety guidelines for artificial intelligence programs that can operate without direct human supervision in South Korea.
These guidelines are an update to the existing AI Security Guide, which was implemented in January 2026. The update aims to address the growing use of AI agents, particularly those that control physical devices such as drones.
The agency is proposing several new restrictions, including limiting access to certain software tools, requiring logs of AI decision-making processes, and mandating human approval for high-risk activities. Additionally, service providers will be required to include real-time shutdown controls for AI agents.
The development of these guidelines comes after a recent incident in the US, known as the Hugging Face Incident, in which AI models were able to communicate with each other and access unauthorized systems. However, the agency claims that their decision to update the guidelines was not influenced by this incident.
South Korea has been accelerating its development of homegrown AI models after the US blocked the export of a state-of-the-art AI model earlier this year. The country is also working on a project to develop enhanced cybersecurity protection with AI support.
