Nearly 40% of South Korean workers are using personal AI tools for work, despite their companies providing official AI solutions. The proportion of employees actually utilizing AI agents in their tasks is below 20%, highlighting a growing phenomenon of 'AI silos' where company and personal AI tools operate separately.
A survey conducted by Samsung SDS, which included 1,750 employees from various domestic companies, revealed that 38% of respondents are using generative AI at their own expense without company approval. This figure significantly exceeds the 24% who use the enterprise version provided by their employers. Meanwhile, the average daily usage of generative AI has increased, reaching 3.25 hours as of May, with the percentage of those using it for more than two hours rising from 18% to 72% in just six months.
AI agents, which are autonomous systems capable of planning and executing tasks, are typically provided by companies due to the high costs associated with server and data integration for individual users. However, only 17% of employees reported using AI agents in their work. The main reasons for not utilizing these tools include a lack of internal guidelines and policies (34%) and the absence of necessary features (33%).
This disparity is referred to in the industry as 'agent sprawl' and 'AI silos.' Agent sprawl describes the uncontrolled proliferation of AI agents across departments and individuals without unified governance, while AI silos refer to the use of different AI systems within departments, leading to disconnected data and workflows.
Among those already using AI agents, 39% have created their own systems on personal computers or closed company servers (41%) or integrated them with internal systems and databases (40%). The use of unauthorized AI tools has also raised security concerns, such as the risk of training these systems on confidential data. About 47% of respondents indicated that their company's AI security guidelines are insufficient, while 54% reported taking personal precautions to mitigate risks.
Conversely, 95% of respondents expressed willingness to actively use a secure and practical enterprise AI agent platform if it were available, suggesting that the avoidance of AI tools is more a result of a lack of company alternatives than a lack of employee initiative.
When selecting AI tools, employees prioritize compliance with corporate security policies and data protection (54%) and compatibility with existing internal systems (51%). If the AI provided by the company fails to meet these criteria, employees are likely to revert to familiar external AI tools, regardless of company efforts to promote adoption.
This issue of AI silos is not limited to private companies; similar challenges are being observed in public institutions and law enforcement agencies, indicating a widespread problem across various sectors.
Even the AI introduced by the Supreme Court has faced rejection from some judges. One judge working in Seoul stated, 'It is at a chatbot level, and the model itself is significantly inferior to global frontier standards, so I do not recommend its use to my colleagues.' Additionally, the 'Police Investigation Support AI (KICS-AI)' introduced last year has been criticized for its poor performance and utility, leading to its neglect internally.
An IT industry insider emphasized, 'The use of personal tools poses a significant risk of leaking company secrets or customer data, so companies must urgently develop a strategy for secure enterprise-level GenAI offerings.'
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.