While an increasing number of companies are seeking to improve operational efficiency through the spread of generative AI, concerns regarding the leakage of internal data and sensitive information remain a stumbling block to adoption. With Theory being selected as a supplier for the government's AI voucher support program, attention is focused on whether the supply of LLM security solutions to alleviate these concerns will begin in earnest.

On the 24th, Teori announced that it has been selected as a supplier for the 2026 AI Voucher Support Program promoted by the Ministry of Science and ICT and the National Information & Communication Technology Promotion Agency (NIPA). The AI Voucher Support Program is a system that provides up to 200 million won per project in purchasing costs to companies that need to adopt AI solutions.

'AlphaPrism,' supplied by Theory, is characterized by a method that identifies sensitive information by analyzing the context of prompts and source code. Designed to detect not only personal information but also internal corporate assets such as cloud authentication keys and secret keys, its key feature is presented as enabling control that reflects actual business context rather than simple keyword blocking.

The response method is also different from a one-sided blocking approach. It applies masking to prompts and the entire file when sensitive information is detected to reduce the risk of information exposure, while allowing the continued use of generative AI for tasks such as summarization and translation. For companies that have hesitated to adopt in-house AI due to security regulations, this can be seen as an approach aiming for a balance between usability and control.

The scope of application is broad, extending to companies across the board seeking to integrate generative AI into their operations. Demand is expected particularly in administrative organizations and development departments utilizing external LLM services, as well as in corporate environments where regulatory compliance is critical. Theori explained that it confirmed the operational feasibility and scalability of AlphaPrism in an enterprise setting by recently implementing it at POSCO DX.

This selection is interpreted as an example demonstrating that security is shifting from a secondary condition to a prerequisite for adoption in the phase of generative AI expansion. Moving forward, it is increasingly likely that data protection and compliance capabilities, alongside model performance, will establish themselves as key criteria for AI adoption in the domestic enterprise market.