Benefits highlighted include reduced need for AI specialists, faster deployment of domain-specific AI, better handling of regulatory changes, and AI that grows with workplace knowledge and operations.

Fujitsu reveals a plan to embed its self-evolving AI technology into its proprietary AI platform, aiming to make it core to in-house AI development and autonomous operation, while collaborating with researchers to enable sovereign AI that can learn on-premises and at the edge and adapt to on-site failures and regulatory changes.

The technology is designed to address societal challenges like personnel shortages and knowledge succession by allowing AI systems to evolve alongside business environments and rules.

Under the One Fujitsu initiative, the approach supports deploying AI inside customer environments, adapting to individual rules and judgment criteria and enabling autonomous business execution with worldwide standardization of IT, data, and processes.

In medical contexts, the approach enables structured extraction from unstructured data such as medical records and test results, improving response accuracy and enabling faster, tailored AI model development without heavy AI-specialist involvement.

Building on that, AI agents can take over tasks previously done by experts in real-time within customer environments by identifying success/failure reasons, extracting knowledge, and updating prompts and evaluation criteria autonomously.

In essence, the technology enables AI agents to continuously learn from outcomes, extract actionable insights, and adjust their behavior autonomously, reducing dependence on human specialists.

The announcement was made in Kawasaki, Japan on May 25, 2026, with Fujitsu asserting standard trademark notices and SDG commitments in the release.

A core feature is that AI agents identify reasons for success and failure, extract actionable knowledge, and autonomously update prompts and evaluation criteria, reducing reliance on human experts for ongoing adjustments.

The technology focuses on two main areas: (1) automated enhancement and continuous evolution of domain-specific large language models via autonomous data selection, learning-condition adjustments, evaluation, and improvement, achieving about a 28-point accuracy gain after optimization; (2) AI agent-powered document search for design specifications in Fujitsu’s health records and local government solutions, improving search strategies and extraction accuracy by learning from past results and corrections.

A practical demonstration with Takane, Fujitsu’s multi-domain AI, showed an average 28-point accuracy improvement after domain-focused optimization, including structured extraction from unstructured medical data.

In the design-specification search use case, AI agents learn from prior results and corrections to broaden search scopes, enhance document extraction, and mimic expert exploration, reducing manual effort and boosting accuracy for health IT systems and local government solutions.