Challenges in moving Japanese companies' use of AI agents from PoC to production have been identified, and data governance is receiving greater attention. The full enforcement of the AI Promotion Act and revision of the Basic Plan for Artificial Intelligence require companies to address data handling and accountability.
Japanese companies using AI are expected to design and disclose their own approaches to proactive use and risk management against the backdrop of the fully enforced AI Promotion Act and the Cabinet Office's guidelines. These frameworks call for explanations of how AI contributes to corporate value and competitive advantage.
This series focuses on governance in the use of AI agents and examines the importance of “control” when delegating decision-making to agents. To ensure agent safety, companies must define for themselves what constitutes safe operation and explain that definition.
In customer inquiry handling, an agent may use an LLM to interpret intent, consult customer-management-system records, check delivery status in an internal system and prepare a response without human intervention. The basis for its authority and the possibility of subsequent verification are at issue.
A joint study by Singapore and South Korea highlighted the risk of data leakage by agents and reported multiple situations in which even non-malicious tasks could not be performed safely. It also identified problems involving access to unnecessary information and inappropriate disclosure of information.
The Basic Plan for Artificial Intelligence, revised in 2026, encourages stronger technical and organisational responses in addition to institutional measures, based on the recognition that AI is evolving into an entity responsible for decision-making and execution.