Kakao to Pre-Screen AI Risks and Establish Safety Evaluation Framework

[Financial News] Kakao is establishing a safety evaluation framework to pre-screen potential risks in artificial intelligence (AI) models. Starting with language models, it will expand the scope to multimodal models and AI agents while jointly developing AI safety evaluation tools.
Kakao said on the 29th that it had signed a strategic memorandum of understanding (MOU) with AISI on the 28th for joint AI safety evaluations and the establishment of an evaluation framework.
As AI technology advances rapidly, the importance of AI safety, including the early identification and verification of potential risks in models, is growing. Last year, Kakao conducted joint safety evaluations with AISI on its in-house language model, 'Kanana-1.5-9.8B.'
Under the agreement, Kakao will provide its AI models and AI agents for evaluation and support the infrastructure needed to conduct evaluations and develop evaluation tools. AISI will be responsible for conducting evaluations and advancing the methodology. The two sides will jointly review the evaluation results, consult on the scope of their disclosure, and cooperate in developing AI safety evaluation tools.
The cooperation will proceed in three stages. First, safety evaluations will be conducted on language models before and after deployment. The scope will then expand to multimodal language models and AI agents. In the final stage, the two sides plan to jointly develop evaluation tools that can efficiently verify the safety of AI models and agents.
Kakao plans to establish a system for proactively identifying and verifying risks in AI models while continuously enhancing its AI safety evaluation capabilities.
Kim Se-woong, Kakao's AI Synergy Performance Leader, said, "We will continue striving to provide AI models and services that prioritize safety and trust alongside technological innovation."
Myung-joo Kim, director of the Artificial Intelligence Safety Institute, said, "Through this cooperation, we will do our utmost to establish an AI safety evaluation framework that meets global standards."
[email protected] Choi Hye-rim Reporter