Artificial intelligence (AI) has long been embedded in financial services, from algorithmic trading to risk management and fraud detection. However, the rise of more advanced AI systems, including generative AI and agentic AI within the broader scope of artificial general intelligence (AGI), is reshaping the landscape. These technologies can generate human-like reasoning, autonomously perform complex tasks, and adapt dynamically to evolving scenarios, pushing AI closer to true decision-making intelligence.

While the United States and China have dominated AI development, Japan is now asserting itself as a key innovator in this next phase of AI evolution. Preferred Networks (PFN) sees itself at the forefront of this transformation, developing generative AI models, high-performance AI chips and agentic AI frameworks designed to work autonomously within business ecosystems. PFN’s research into AGI is particularly significant as it aims to bridge the gap between narrow AI applications and more generalised, human-like intelligence.

At the TAB Global Japan Innovation Study Tour 2025, Kageatsu Ueno, vice president of global AI business development at PFN, led discussions on how banks can integrate these advanced AI capabilities into their operations. Adiyan Mujibiya, vice president of preferred AI products, and Karim Hamzaoui, engineering manager expanded on PFN’s AI-driven solutions, research directions and strategic collaborations. Their insights reflect PFN’s global outlook, with Mujibiya, an AI leader who hails from Indonesia, and Hamzaoui, who brings AI expertise from Morocco, offering diverse perspectives to PFN’s innovation strategy.  The session underscored the shift from rule-based automation to self-learning, adaptable AI systems, highlighting how PFN’s innovations may impact the future of financial services.

A full-stack approach to AGI and agentic AI

Unlike many AI firms that focus solely on traditional deep learning, PFN is actively developing AI systems that can reason, plan and operate autonomously - key pillars of AGI and agentic AI. PFN claims its proprietary next generation AI chips are designed to be more energy efficient to support agentic AI systems, enabling faster decision-making and real-time adaptation to new environments. These AI agents can handle complex workflows with minimal human intervention, a capability that could revolutionise banking operations, compliance automation and personalised financial advisory services.

“We are moving beyond pattern recognition towards AI systems that can act independently and optimise decision-making,” Ueno explained.

Beyond hardware, PFN has built a robust agentic AI infrastructure that integrates multi-modal AI processing, allowing AI systems to interpret text, speech and images in a business context. These capabilities are critical for organisations looking to automate regulatory compliance, customer service interactions and AI-powered risk assessment models.

PFN’s proprietary large language model for Japan

A key feature of PFN’s AI ecosystem is Plamo, a generative AI large language model (LLM) developed specifically for Japan’s language and business landscape. Unlike conventional AI models, which rely on pre-programmed responses, Plamo has been designed to achieve high-context accuracy, allowing for improved comprehension of Japanese language nuances, cultural references and industry-specific terminology. The next iteration of Plamo will also be developed for English.

PFN believes Plamo’s training methodology enables it to surpass GPT-4 and Claude in Japanese language benchmarks, a capability that positions it as a viable alternative to foreign-developed LLMs. “Plamo is a model that understands Japan’s unique linguistic and business context better than generic AI models,” Mujibiya stated.

While Plamo is not explicitly optimised for finance, its ability to process high-context Japanese data makes it adaptable to multiple industry applications, including financial services. PFN is also positioning Plamo for future expansion into additional languages, reinforcing its commitment to advancing AI-driven automation beyond Japan.

SBI investment strengthens PFN’s AI capabilities

PFN recently secured investment from SBI Holdings to accelerate its AI and semiconductor development. This funding underscores SBI’s confidence in PFN’s AI-driven innovations and supports PFN’s ambition to challenge global leaders in AI hardware, particularly in energy-efficient AI chips.

The investment also strengthens PFN’s ability to expand its AI applications beyond Japan, providing additional resources for scaling its LLM infrastructure, AGI research and agentic AI development. SBI’s involvement is a testament to PFN’s growing role in Japan’s AI ecosystem and its potential to shape AI advancements globally.

SBI Holdings has established a robust international investment portfolio, particularly in the financial sector. Notable investments include a 19.9% stake in TPBank in Vietnam and a significant investment in Ripple Labs, the company behind the cryptocurrency XRP. Additionally, SBI is reportedly set to acquire a majority stake of more than 70% in the German embedded finance platform Solaris. ​

The investment in PFN also strengthens the company's ability to expand its AI applications beyond Japan, providing additional resources for scaling its LLM infrastructure, AGI research and agentic AI development. SBI’s involvement is a testament to PFN’s growing role in Japan’s AI ecosystem and its potential to shape AI advancements globally.​

This partnership reflects a broader trend where financial institutions recognise that AI leadership requires multinational partnerships and global data-driven strategies. SBI’s support enables PFN to accelerate the development of AI systems that are not just Japan-centric but globally competitive, addressing challenges in AI ethics, security and regulatory compliance that extend beyond a single market.

Data differentiation, RAG and AI benchmarking

Mujibiya highlighted that an organisation’s unique data is the key differentiator in its use of generative AI. Rather than relying solely on pre-trained AI models, businesses—particularly banks—must leverage their proprietary data to enhance AI accuracy and relevance.

“Because the data comes from the market, and the market is unique, how you strategise your customer is unique and therefore what comes from it is unique,” explained Mujibiya, on why data becomes the differentiator.

To optimise AI-driven insights, PFN promotes the use of retrieval-augmented generation (RAG). This technique allows generative AI models to reference live, organisation-specific data in real-time rather than relying on static training datasets. By integrating RAG-based AI deployments, banks can significantly improve the contextual relevance of AI-generated outputs while maintaining competitive differentiation.

AI benchmarking was another key theme. PFN systematically benchmarks AI models against global alternatives, including GPT-4 and Claude, to ensure that Plamo maintains a superior understanding of Japan-specific language, business dynamics and regulatory frameworks. PFN’s approach ensures that its AI models remain highly competitive and adaptable to evolving business requirements.

Challenges and limitations of AI adoption in financial services

While AI offers transformative benefits, banks face several practical challenges in adopting AGI:

• Regulatory compliance: AI must comply with stringent financial regulations, and ensuring transparency in AI-driven decisions remains a key challenge.
• Data security risks: The use of AI models that process live transaction data raises concerns about customer privacy and cybersecurity threats.
• Computational costs: Large-scale AI models require significant computing power, making on-premise or sovereign AI solutions more viable for data-sensitive financial institutions.

PFN said one way to address these concerns is by adopting secure, on-premise AI deployment options, ensuring banks maintain control over data governance and regulatory compliance.

How banks can optimise their generative AI strategy

PFN’s discussions at the study tour outlined several key strategies for banks looking to enhance their generative AI deployment:

• Invest in data readiness: Ensure clean, structured, and well-curated data to maximise AI performance.
• Use RAG: Allows generative AI models to reference live, bank-specific data rather than relying solely on pre-trained knowledge.
• Prioritise AI security and compliance: Work with AI providers that offer on-premise deployment and sovereign AI solutions to protect sensitive financial data.
• Leverage agentic AI models: These models can autonomously refine financial insights and decision-making strategies over time.

How banks should embrace the next phase of AI

The conversations with PFN at the Japan Innovation Study Tour made one thing clear: generative AI, AGI and agentic AI are ushering in a new era of AI-driven intelligence. Organisations that remain reliant on traditional AI models risk falling behind as AI systems become increasingly autonomous and strategic.

PFN’s generative AI capabilities, from Plamo to autonomous agentic AI systems, illustrate how businesses can move beyond rule-based automation towards self-learning, decision-making AI frameworks. However, success in this space depends on how organisations approach AI integration.

With data differentiation, RAG integration and AI benchmarking now central to AI strategy, banks and financial institutions must ensure they are not just AI adopters but AI leaders, shaping how intelligence is applied to financial services.