en.Wedoany.com Reported - LG AI Research unveiled a legal risk management strategy for AI training data and released the beta version of "Exaone Nexus," a data source and license tracking analysis platform, at a seminar titled "Invisible Legal Risks of AI Training Data" held in Seoul.
As artificial intelligence regulations tighten, managing legal risks in training data has become increasingly important. Lee Hwa-young, Executive Director of LG AI Research, stated at the seminar that over 99% of training data for generative AI models comes from open datasets, while manually created or licensed data typically accounts for only 1% to 2%. He noted that open datasets often contain multiple nested sub-datasets, which may include data not permitted for commercial use, making it difficult to determine actual usability based solely on top-level licenses. Data with restricted commercial use or personal information issues is highly likely to be mixed in, creating legal risks at every stage of copying, modification, and distribution.
LG AI Research has been collaborating with law firm Yulchon for three years to address data legal issues, and the two parties jointly unveiled the "Exaone Nexus" platform. This platform aims to assess the legal risks of open datasets, supporting the development and adoption of AI models, with a planned commercial launch in September this year.
"Exaone Nexus" can detect sub-datasets within a dataset, confirming actual data sources and license conditions. Its feature lies not in merely viewing top-level dataset information, but in tracing the entire data structure to check for license contamination. The platform incorporates three AI agents: one responsible for finding data and license documents, another for evaluating 18 items (including commercial usability, personal information issues, usage region and time restrictions, etc.), and the remaining agent for verifying analysis results.
LG AI Research classifies datasets into three grades—A, B, and C—based on evaluation results. Data with extremely low legal risk is classified as Grade A, data with potential disputes is managed as Grade B, and data with high legal risk is classified as Grade C and excluded from the training process. AI models are also managed using the A, B, C grading system, as legal disputes or regulatory changes may still occur after training, requiring continuous risk assessment.
Executive Director Lee Hwa-young stated that not only AI companies but also enterprises adopting AI and investment firms can utilize the platform. Companies adopting AI can review the legality of model training data, while investment firms can pre-check the legal risks of their portfolio companies. Im Woo-hyung, President of LG AI Research, emphasized that AI competition is shifting from a performance-oriented focus to one centered on trust and responsibility, making training data source and license management an urgent issue. "Exaone Nexus" can track training data sources throughout the process and verify license compliance, helping companies manage data risks more clearly. Kang Seok-hoon, Lead Attorney at law firm Yulchon, pointed out that the proliferation of generative AI has increased legal risks such as copyright infringement, data usage rights, and disputes between rights holders and AI companies, requiring companies to simultaneously examine data sources and rights relationships, licensing systems, development process record management and internal controls, and dispute response systems.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
