Kim Hyungjun, BHSN Chief AI Officer Interview
Practical Automation with Legal-Specialized Model 'Allyby Astro'
"Standards Needed to Define AI's Responsibilities and Roles"
From drafting complex contract templates to summarizing case law, artificial intelligence (AI) is becoming a practical tool for legal document work. BHSN, which has developed its own large language model (LLM) specialized for the legal field, is leading the legal AI market by expanding real-world automation use cases for legal tasks, particularly among major domestic and international corporations and public institutions.
Kim Hyungjun, Chief AI Officer (CAIO) of BHSN, said in an interview with Asia Economy on August 5, "Legal AI should not replace lawyers, but rather serve as a practical tool that helps lawyers make faster and more accurate decisions," adding, "We are focusing on developing AI that actually drafts contracts, summarizes case law, and reviews regulations."
'Allyby Astro,' developed by BHSN, is a proprietary LLM designed to handle complex processes such as drafting contract templates, summarizing case law, and searching statutes. It is optimized for understanding Korean sentence structures and is also being applied in on-premises environments for some companies in Japanese and Chinese. Its strength lies in being built as a pre-trained model based on continuous pre-training (CPT), going beyond simply fine-tuning a general-purpose LLM.
Kim, the CAIO, explained, "Compared to GPT-4, it operates about three times faster and achieves over 10% higher accuracy. We are also applying multimodal capabilities so it can process unstructured information such as images and tables," adding, "Whereas existing LLMs had speed limitations due to heavy computational loads when searching through large volumes of documents, Astro has been streamlined to a 2-4 bit model structure and optimized storage methods, improving both search speed and accuracy."
In particular, it is designed as a non-learning structure that does not train on customer data, earning trust from large enterprises that handle sensitive legal documents. Major companies such as CJ CheilJedang and Aekyung Chemical have adopted BHSN's legal AI solutions, automating contract review and advisory requests not only in their legal teams but also in HR, finance, and sales departments.
Although this technology is already replacing or supplementing work in the field, there are concerns that domestic regulations remain at an early stage. The 'Legal Tech Industry Activation Guidelines' recently announced by the Ministry of Justice have yet to include specific content regarding AI utilization. Kim, the CAIO, stated, "AI is merely a tool to supplement the judgment of legal professionals," and emphasized, "Institutional improvements are needed to define its role and scope of responsibility."
In Japan, the Ministry of Justice is proactively establishing institutional frameworks and policy guidelines for the entire legal tech sector, bringing the market into the regulatory fold. In particular, Japan is accelerating market adoption by clearly defining usage standards and responsibility for AI-based contract review systems. Kim, the CAIO, explained, "Japan is much larger than Korea in terms of population and market size, and the Ministry of Justice has a clear direction regarding AI utilization," adding, "There is active demand and investment in legal tech, so we are open to technological partnerships with local companies and plan to expand our market presence."
This year, BHSN is expanding customer-tailored adoption of its subscription-based legal AI service 'Allyby' and responding flexibly to various implementation methods. Kim, the CAIO, said, "As demand becomes more segmented?not only for SaaS models but also for on-premises types or customers seeking only specific functions?we have built agile capabilities to provide customized solutions," adding, "Our goal is to continue establishing ourselves as a trusted legal AI company that customers can rely on."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
