LG’s EXAONE Nexus assesses legal risks in AI training data

LG AI Research describes EXAONE Nexus as an AI compliance agent that automatically analyzes legal risks in large AI training datasets.[2] Those risks include copyright, licensing and privacy issues.[3] LG says the system traces data to its source to identify legal issues that may arise when original material is redistributed or combined with other data.[3] The company said it developed the system in response to concerns that copyright disputes over training data were raising legal risks for companies and undermining trust in AI.[2] This positions Nexus as a risk-assessment tool; the evidence does not say it settles disputes or guarantees that a dataset is legally cleared.

LG AI Research reported that EXAONE Nexus analyzes one dataset 45 times faster than a legal expert.[3] The company also said the system achieves 26% higher accuracy and costs one-thousandth as much as conventional methods.[3] The supplied information does not specify the accuracy comparison baseline or the testing conditions behind these figures. The numbers may help teams assess the potential value of automated screening, but they should be treated as LG AI Research’s reported results rather than as independently established outcomes.

LG AI Research said it built South Korean and European Union AI regulations into EXAONE Nexus.[3] The company’s compliance framework incorporates South Korea’s AI Basic Act and the EU AI Act.[4] LG said it worked with about 40 lawyers from major domestic and international law firms to verify the legal validity of the system’s analysis.[3] Its assessment framework uses 18 criteria to sort results into seven grades, from A-1, the safest, to C-2, the highest risk.[3] These details describe the framework and review process, but do not show how every possible legal issue is handled.

LG has adopted EXAONE Nexus as a mandatory verification step in developing its own AI models.[3] LG AI Research said it uses the system throughout the AI-development process, from data collection through model development and service launch.[3] The company released risk-assessment results for about 20,000 open datasets for research purposes.[3] It plans to expand supply to overseas big-tech companies and law firms, but has not disclosed the supply timing or business scale.[3][5] LG AI Research also says the system is based on its EXAONE 3.5 model and includes navigation, question-answering and scoring modules.[6]