It Even Operates the Mouse and Keyboard for Users
Bans Spread amid Security Concerns over AI Agents
Some Praise Its High Practical Utility
"We Need to Build and Test as Much as We Can"
The use of the artificial intelligence (AI) agent (assistant) “Opeunkeullo,” which can operate a PC and perform work on behalf of users, is increasing. While some say it can practically replace human work, more IT companies are banning its use due to rising security concerns.
According to the IT industry on February 9, companies such as Naver, Kakao, and Danggeun Market recently announced that they are banning in-house use of Opeunkeullo, an open AI agent. Kakao has prohibited the use of Opeunkeullo on its internal network and work devices in order to protect internal information assets, and Naver and Danggeun Market have likewise notified employees that use of Opeunkeullo inside the company is restricted. This is the first time since last year’s case of the Chinese generative AI “DeepSeek” that Korean IT companies have imposed such broad restrictions on the use of an AI service.
Opeunkeullo is an open-source AI agent developed by Austrian developer Peter Steinberger. The AI can effectively perform work by checking the computer screen and operating the mouse and keyboard on behalf of the user. It can write emails or documents and manage schedules for the user, and it can even take over tasks that require direct handling of the PC, such as organizing files on the PC and running programs. The agent was initially launched under the name “Claudebot,” but after Anthropic, which operates the generative AI service “Claude,” raised a trademark issue, the name was changed first to “Maltbot” and then to “Opeunkeullo.”
Overseas as well, there are moves to block the use of Opeunkeullo and voices raising security concerns. According to foreign media reports, China’s Ministry of Industry and Information Technology (MIIT) warned on February 5 (local time) that if Opeunkeullo is configured inappropriately, it could become a pathway for cyberattacks and data leaks. Microsoft (MS) in the United States, through its AI safety team, has also officially expressed concern that “Opeunkeullo is still too vulnerable in terms of security to be used in enterprise environments.”
Such concerns are being raised because, for the AI agent Opeunkeullo to operate, it must access and then utilize a wide range of information. It can autonomously perform work on behalf of users, but in the course of operation it ends up using trade secrets or sensitive personal information. In this process, there is no way to completely rule out the possibility that Opeunkeullo could malfunction or that information could be leaked due to security flaws. Since Opeunkeullo itself is an open-source project created by an individual, its limitations in terms of security are also clear.
In fact, there have been reports of vulnerabilities in Opeunkeullo where users’ application programming interface (API) keys were exposed or where, through “indirect prompt injection” attacks, the AI siphoned off sensitive personal and financial information.
Even so, some say Opeunkeullo could help popularize “agentic AI” systems that can practically replace human work. Kim Sunghoon, CEO of Upstage, wrote on his social media (SNS), “The emotions I felt last week while using Opeunkeullo, Maltbook, and Botmadang were surprise, fear, and excitement,” adding, “The singularity of AI is approaching. We need to approach it cautiously, but at the same time test and build as much as we can.”
Apple’s desktop “Mac mini,” which is regarded as the optimal device for running Opeunkeullo thanks to its low price and excellent AI computing performance, is in short supply at home and abroad. Major domestic e-commerce platforms are out of stock of the Mac mini, making it impossible to purchase, and Apple’s official website is indicating that delivery will take more than one month.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
