Fujitsu has announced the launch of a dedicated AI platform designed to allow enterprises to autonomously manage the full generative AI lifecycle, from model development and operation through to incremental learning and continuous improvement of AI models and agents.
The platform will be rolled out sequentially across Japan and Europe, with preliminary trial registrations scheduled to open in February 2026. Fujitsu said it will subsequently accelerate deployment across all AI scales, ranging from large enterprise environments to edge and physical AI use cases.
The new platform is intended to support organisations seeking sovereign AI capabilities, particularly those operating in regulated or security-sensitive environments where protection of confidential data and direct control over AI models are critical. Fujitsu noted that many enterprises face challenges deploying in-house AI platforms, including shortages of skilled AI engineers, complex operational management, rising compute demands and evolving security threats.
To address these issues, Fujitsu’s platform can be deployed in dedicated, closed environments, including on-premise installations or within Fujitsu-operated data centres. The solution will be delivered via the Private AI Platform on PRIMERGY and Private GPT, integrating Fujitsu’s Kozuchi AI technologies into a single package aimed at reducing deployment complexity and enabling adoption without deep specialist expertise.
Security and trust are central to the platform’s design. Fujitsu said the system incorporates vulnerability scanning covering more than 7,700 known vulnerabilities, including Fujitsu-defined categories, alongside guardrail technologies to detect and suppress malicious attacks such as prompt injection, inappropriate outputs and unexpected AI behaviour both before execution and at runtime. Automated rule generation and application are designed to maintain stable and reliable AI operations, even for non-specialist users. The company also plans further enhancements to reduce hallucinations and improve output reliability.
At the model layer, the platform is built around Fujitsu’s Takane large language model, optimised for high-precision Japanese language performance and image analysis. Customers can fine-tune models in-house to continuously improve performance for specific business needs. Fujitsu said model lightweighting technologies can reduce memory consumption by up to 94 per cent, while integrated quantisation techniques are designed to lower operational costs and improve performance efficiency.
The platform also includes efficiency tools for AI agent development, offering low-code and no-code capabilities to accelerate deployment by on-site teams. Support for Model Context Protocol and inter-agent communication allows integration with existing systems and cooperative operation between multiple AI agents. Fujitsu said it is also exploring options to containerise and deliver its proprietary technologies as AI agents on demand.
Fujitsu said the platform reflects its broader strategy to continuously evolve generative AI in line with changing business requirements, enabling safe, reliable use across industries and supporting long-term customer transformation and growth.
