[비즈한국] Major domestic IT companies that champion artificial intelligence (AI) have recently issued a collective 'usage ban' to their employees regarding a specific AI technology. The target is 'OpenClaw,' an AI agent that performs tasks directly on a computer that were previously done by humans. Companies including Naver035420, Kakao035720, and Danggeun have recently restricted the use of OpenClaw or blocked access to it for their employees. What is the reason these companies are drawing a line with OpenClaw?
OpenClaw is an autonomous agent that can read and write files, operate a mouse and keyboard, and execute scripts on behalf of the user. OpenClaw, which can run in a local environment, is gaining attention as an 'AI that replaces humans' because it can link with messengers like WhatsApp and iMessage to perform tasks such as booking flights, managing schedules, and sending messages. Public interest has grown rapidly, especially with the emergence of 'Moltbook,' which is promoted as the dedicated social media platform for OpenClaw. Developer Peter Steinberger stated that OpenClaw gained 2 million visitors within a week of its release.
However, domestic IT companies saw the potential risks before the convenience. Kakao recently announced internally, "To protect company information assets, we are restricting the use of the AI agent OpenClaw (formerly Claudebot/Moltbot) on corporate networks and work devices." Naver also issued a ban, and Danggeun blocked access to OpenClaw and Moltbook. The decision was driven by the judgment that current security systems are insufficient to control agent-type AI that can access internal corporate systems and data. This is the first time since the Chinese AI model 'DeepSeek' early last year that these companies have explicitly restricted the use of a specific AI.
Similar trends are observed overseas. According to Reuters and other reports, the Chinese Ministry of Industry and Information Technology warned on the 5th that "if not configured properly, OpenClaw could pose serious security risks," stating that users and organizations could be exposed to cyberattacks and data leaks. While not a total ban, they have recommended that organizations adopting OpenClaw check for public network exposure and strengthen identity authentication and access control. This is because instances of some users employing OpenClaw without even basic security settings have been confirmed.
The security industry also points to the 'operating method' rather than the technology itself as the problem. Global network security firm Cisco recently criticized in a report that "OpenClaw is an innovative tool that personal AI assistants have long aimed for, but from a security perspective, it is close to a 'nightmare'." The analysis suggests that an AI agent with system access privileges could create 'covert leakage channels' that are difficult for existing Data Loss Prevention (DLP) solutions or endpoint security tools to detect.
They also noted that because the prompt itself functions as an execution command, it is difficult for existing security tools to identify them, and installing externally distributed 'Skills' without verification could amplify supply chain risks. OpenClaw's skills are installed as local file packages, leading to the analysis that the possibility of malicious code infiltration persists and that as the skill ecosystem grows, risks will also be amplified.
This measure by the domestic industry is interpreted as a judgment that the corporate environment does not yet meet the security requirements needed for agent-type AI. As it is designed based on the premise of file access and integration with external services, there is a high concern that it will act more as a 'security vulnerability' than a 'productivity tool' in a corporate environment where many personnel share systems.
OpenClaw differs from existing generative AI in that it is an 'agent-type AI' that goes beyond conversational AI, which merely answers questions, to directly intervene in the user's computer environment to perform tasks. It is also mentioned that basic security for OpenClaw is currently closer to an optional feature, and that control models suitable for corporate environments have not yet been sufficiently established.
The overseas IT media outlet TechCrunch evaluated OpenClaw as "the most direct implementation of the automation and autonomy that personal AI assistants have long pursued," but pointed out that "as it is premised on system access, its adoption may face restrictions in corporate environments unless authorization management and security models are designed alongside it."