Kakao will work with the AI Safety Institute to build an evaluation system to verify the safety of its AI models and agents.

Kakao said on Monday it signed a memorandum of understanding on Sept. 28 with the AI Safety Institute for joint AI safety assessments and the development of an evaluation framework.

The partners plan to gradually expand the scope of evaluations from language models to multimodal language models and AI agents. In the long term, they will also develop a joint evaluation tool to verify the safety of models and agents.

Kakao will provide its AI models and agents to be evaluated and support the infrastructure needed to conduct assessments and develop tools. The AI Safety Institute will handle the evaluations and advance the methodology.

The assessment results will be reviewed jointly, and the companies will also discuss the scope of any external disclosure.

The cooperation will proceed in 3 stages. First, the partners will conduct pre- and post-deployment safety evaluations of language models and then expand the targets to multimodal language models and AI agents. In the final stage, they will jointly develop a safety evaluation tool that can be used repeatedly.

Kakao previously conducted a joint safety assessment last year with the AI Safety Institute on its in-house language model, Kanana-1.5-9.8B.

Se-woong Kim (김세웅), Kakao's AI Synergy Performance Leader, said, "Through this cooperation, we will be able to apply a more objective and rigorous safety verification system to Kakao AI models." He added, "We will continue to advance AI models and services based on safety and trust."

Myung-joo Kim (김명주), head of the AI Safety Institute, said, "As AI technology becomes more advanced, it is important to have a system that proactively identifies and evaluates potential risks." He added, "Through cooperation with Kakao, we will build an AI safety evaluation system that meets global standards."