In South Korea, major IT companies including Naver, Kakao, and Carrot have imposed a ban on the usage of the open-source AI agent Open Clo internally to address security concerns. Known for its ability to manipulate tasks through a computer screen, Open Clo has been likened to a real-life version of Jarvis from Iron Man, but the potential security vulnerabilities it presents have prompted companies to take action. Kakao confirmed that they conducted a security briefing regarding Open Clo for their developers and restricted its use on corporate networks and devices in order to protect their information assets.

Open Clo operates continuously as long as the device is powered on, allowing users to issue commands via messaging apps like Telegram, making it a convenient tool for everyday tasks. However, the extensive permissions held by this AI agent raise alarms about potential breaches in corporate security systems. Justin Capos, a professor at NYU, highlighted the risks by comparing granting computer control to an AI agent to giving a knife to a child, stressing the importance of caution when integrating such technology into business operations.

The concerns are underscored by findings from a malware tracking group, which identified 386 malicious skills that could be exploited by AI agents, posing significant risks of information theft under the guise of useful functionalities. Jason Meller, Vice President of Product at the security solutions company OnePassword, has also warned that running Open Clo on corporate devices should be viewed as a potential security incident and advised immediate contact with the security team. Reports indicate that even the Chinese Ministry of Industry and Information Technology has acknowledged the severe security risks associated with improper configurations of Open Clo, emphasizing the growing global awareness of the implications of AI in the corporate landscape.