The framework focuses on ensuring verifiable information sources and enabling AI systems to stop generating answers when uncertainty is high.

This initiative enhances the reliability of AI-generated answers by enabling systems to provide evidence-based responses and to refrain from answering when uncertainty is high—addressing persistent challenges in the business use of rapidly advancing generative AI. Mitani Sangyo is building intellectual property to contribute to global discussions on AI safety, positioning this framework as an AI governance technology originating in Japan.

Why it matters

Generative AI has advanced rapidly through reasoning models improved with advances in inference models, retrieval-augmented generation (RAG), and AI agents. However, even in these systems, the risk of plausible but incorrect responses—so-called hallucinations—remains a challenge, particularly in mission-critical business settings.

When utilizing AI for critical business tasks, it is essential not only to require correct answers, but also to ensure verifiability and reliability. For example, being able to confirm the basis of information and stopping answers in case of uncertainty. In sectors requiring exceptional accuracy and explainability, such as finance, legal, manufacturing, and public services, preventing incorrect outputs takes precedence over conversational fluency. This necessitates explainable capabilities, such as verifying the data provenance and timeframe (freshness) of information, and fail-safe mechanisms that detect uncertain responses to hold them for human intervention.

To address these challenges, Mitani Sangyo’s "AI Reliability Governance Framework" provides step-by-step management through three checkpoints—Input, Process, and Output—that control how AI systems generate responses and ensure information is traceable, consistent, and safe before any answer is delivered.

Press release: https://www.acnnewswire.com/docs/files/20260416_EN.pdf