[Edaily Reporter Lee So-Hyun ] Kakao ( Kakao(035720)) is establishing a safety assessment system to proactively verify the risks associated with its proprietary artificial intelligence (AI) models and AI agents. Starting with pre- and post-deployment evaluations of language models, the company will expand its verification scope to include multimodal models and AI agents, and in the long term, it plans to jointly develop safety assessment tools.

Kakao announced on the 29th that it had signed a strategic memorandum of understanding (MOU) with the Artificial Intelligence Safety Research Institute on the 28th to conduct joint AI safety evaluations and establish an evaluation framework.

This move comes as the importance of safety assessments—which identify and verify potential risks in models in advance—has grown alongside the rapid advancement of AI technology.

Last year, Kakao conducted a joint safety assessment with the AI Safety Research Institute on its proprietary language model, “Kanana-1.5-9.8B.” Kakao explained that the assessment confirmed a high level of safety compared to overseas models of a similar scale.

Expanding Evaluation from Language Models to

Multimodal and

Agent Models

Through this agreement, the two companies will expand their collaboration on AI safety assessments in three phases.

First, they will conduct safety assessments of language models both before and after service deployment. Subsequently, they plan to broaden the scope of evaluation to include multimodal language models and Agent AI. In the final phase, they will jointly develop evaluation tools that can more efficiently verify the safety of AI models and Agents AI.

Kakao will provide its own AI models and Agent AI for evaluation and support the infrastructure needed to conduct the evaluations and develop the tools.

The AI Safety Research Institute will be responsible for conducting safety assessments and refining assessment methodologies. The two companies will jointly review the assessment results and discuss the scope of disclosure, while also collaborating on the development of assessment tools.

“Establishing a System for Proactive Identification of AI Risks”

This collaboration focuses on creating a system to proactively identify and verify the potential risks of AI models and on continuously enhancing safety evaluation capabilities.

Kakao plans to secure world-class AI safety standards and enhance trust in its AI services through collaboration with a domestic research institute specializing in AI safety.

Kim Se-woong, Kakao’s AI Synergy Performance Leader, stated, “Through this collaboration, Kakao’s AI models have been equipped with an even more objective and rigorous safety verification system,” adding, “We will strive to provide AI models and services that prioritize safety and trust alongside technological innovation.”

Kim Myung-ju, Director of the Artificial Intelligence Safety Research Institute, remarked, “As AI technology becomes more sophisticated, it is crucial to have a system that proactively identifies and assesses potential risks,” adding, “We will work with Kakao to establish an AI safety evaluation system that meets global standards.”

Meanwhile, to strengthen AI safety, Kakao has signed a business agreement with KTCorporation Cloud to build a secure AI ecosystem and is also operating an advisory group called the “AI & Technology Ethics Subcommittee,” composed of external experts.