Japan Eyes AI Audit, Certification Framework

Yomiuri Shimbun file photo
Prime Minister Fumio Kishida and other Group of Seven leaders pose ahead of a summit meeting in Hiroshima on May 19.

A third-party audit and certification system for generative artificial intelligence is among policies being considered by the government to ensure AI firms comply with relevant laws and regulations, according to a draft outline of measures seen by The Yomiuri Shimbun.

The draft stresses the need to improve transparency in the AI sector amid the rapid spread of generative artificial intelligence models that can create text and images based on data from the internet, among other sources.

The government’s AI Strategy Team, comprising officials from relevant ministries and agencies, will discuss details of the measures, government sources said.

According to the draft outline, three pillars of the measures include efforts to ensure thorough compliance with relevant laws, regulations and guidelines; the adoption of a risk-based approach, in which regulations are created according to the level of risk; efforts to address issues through technological development.

A third-party audit and certification framework for AI system developers and operators is being considered to ensure the quality of models, the content of data used to train the models, and the handling of data, according to the draft.

The draft also states the need for user guidelines to raise awareness of issues concerning confidentiality and inaccurate information.

A risk-based approach is likely to be one of the foundations of the European Union’s legal framework on AI. It will involve classifying AI systems according to the level of risk — such as the risk of causing mental or physical harm — and varying the severity of regulations according to the classified risks.

A similar approach will be adopted in Japan, and existing laws, regulations, and guidelines will be reviewed, according to the draft.

Technology to recognize and reduce risks associated with AI systems is an area tipped for future development in the draft. Such efforts could include tools to detect inaccurate information on social media.

The Group of Seven industrialized nations agreed to establish the “Hiroshima AI process” to discuss generative AI at the G7 summit in Hiroshima in May. The bloc is conducting discussions among member states and is expected to announce its position regarding a legal framework for generative AI.

The Japanese government plans to gather information on the situation in other G7 member countries while working out its own framework.