Fujitsu's Multi-AI Agent Revolution: How Companies Can Finally Share AI Without Sharing Secrets

 

The biggest barrier to AI in supply chains isn't technology—it's trust. Fujitsu just solved it.

For years, supply chain executives have faced an impossible choice: share sensitive business data to enable AI collaboration, or protect their competitive secrets and miss out on efficiency gains. It's the digital equivalent of asking rival companies to open their vaults to each other—and understandably, most have refused.

But on December 1, 2025, Fujitsu announced a breakthrough that could change everything. The Japanese tech giant unveiled a multi-AI agent collaboration technology that allows AI systems from different companies to work together seamlessly—without any company having to expose its confidential data to competitors.

Think of it as AI agents negotiating in a dark room, learning about each other through conversation rather than by reading each other's files.

The Problem: AI Works Best When It Knows Everything (But Companies Won't Share)

Modern supply chains are massively complex networks spanning manufacturers, suppliers, logistics providers, pharmaceutical companies, and retailers—often across multiple countries. When disruptions hit—a sudden spike in demand, a natural disaster, or a port closure—the ripple effects cascade through every connected company.

AI excels at optimizing these networks, but there's a catch: traditional AI systems need access to comprehensive data from all parties involved. A manufacturer's AI needs to know about their supplier's inventory levels, production capacity, and delivery schedules. The supplier needs visibility into the manufacturer's demand forecasts. Logistics companies need information from both sides to optimize routing.

The problem? No company wants to hand over this information. It represents competitive intelligence, trade secrets, and strategic advantages. Research from the World Economic Forum highlights that organizational hesitation over data-sharing remains one of the biggest obstacles to supply chain resilience, even when the benefits of collaboration are clear.

As Devavrat Bapat, Head of AI/ML data products at Cisco, explains in a recent CIO article, "What's missing are techniques that allow organizations to share some part of their data in full confidence that they haven't given away too much."

Until now, that missing piece has kept supply chain AI from reaching its full potential.

Fujitsu's Solution: AI Agents That Learn Without Looking

Fujitsu's innovation centers on a radically different approach: instead of pooling data in a central repository, their technology enables AI agents from different companies to approximate each other's characteristics through a negotiation-based dialogue.

Here's how it works in practice:

1. Knowledge Distillation Without Data Sharing

During setup, Fujitsu's system uses a technique called knowledge distillation—a deep learning method where a "student model" learns from multiple "teacher models" without accessing their underlying data. The AI agents share insights about their operational patterns and capabilities without revealing the raw information behind those patterns.

It's similar to how chess players might discuss strategy without revealing their specific game plans for an upcoming tournament.

2. Secure Gateway Communication

Each company's AI agent operates behind a secure gateway built on distributed AI learning technology. This gateway acts as a smart filter, protecting confidential information while enabling meaningful collaboration. During operation, the system continuously monitors communications to detect malicious queries and prevent the inference of confidential data.

The technology simulates AI agent behavior and responses repeatedly, updating and sharing information in formats that can't be reverse-engineered to expose sensitive details.

3. Negotiation-Based Optimization

Once the AI agents understand each other's approximate characteristics, a "proposing agent" can identify optimal solutions for the entire supply chain network. The agents essentially negotiate outcomes that benefit all parties without any single party seeing inside the others' operations.

For example, in optimizing logistics routes, one company's AI might propose a delivery schedule. Other companies' AI agents respond with their constraints and preferences. Through iterative exchanges, they converge on a solution that minimizes overall costs and delays—all without anyone sharing proprietary shipment data, customer lists, or pricing structures.

Real-World Impact: 30% Cost Reduction in Initial Trials

Theory is one thing; results are another. Fujitsu partnered with Rohto Pharmaceutical and the Institute of Science Tokyo to test this technology on a virtual supply chain, and the initial findings are striking.

The trials demonstrated a potential reduction of up to 30% in transportation costs through optimized logistics routes and schedules—achieved without Rohto having to expose sensitive pharmaceutical supply chain data to external partners.

Starting in January 2026, Fujitsu will launch more comprehensive field trials using Rohto's actual supply chain, running through March 2027. These real-world tests will simulate emergencies like sudden demand shifts and natural disasters to evaluate how quickly the collaborative AI system can facilitate recovery.

Professor Katsuki Fujisawa from the Institute of Science Tokyo notes that this approach enables optimization across "the entire industrial value chain" while maintaining the security and autonomy each organization requires.

Why This Matters Beyond Supply Chains

While Fujitsu is initially targeting supply chain optimization through its Uvance business model's Dynamic Supply Chain services (planned for the end of fiscal 2026), the implications stretch far beyond logistics:

Healthcare: Hospitals could coordinate patient care and resource allocation with pharmaceutical companies and medical device manufacturers without compromising patient privacy or proprietary treatment protocols.

Manufacturing: Factories across industries could synchronize production schedules with suppliers and distributors to minimize waste and maximize efficiency—without revealing production capacities or client lists.

Financial Services: Banks and fintech companies could collaborate on fraud detection and risk assessment while maintaining strict data privacy and regulatory compliance.

Energy Grids: Utility companies could optimize power distribution and load balancing across interconnected grids without exposing customer usage patterns or infrastructure vulnerabilities.

The common thread? These are all industries where collaboration yields massive benefits, but data sensitivity has historically prevented it.

The Broader Context: Japan's AI Competitiveness Strategy

Fujitsu's announcement didn't happen in isolation. The company explicitly positioned this technology as a contribution to the Council on Competitiveness-Nippon (COCN), an industry initiative aimed at strengthening Japan's industrial competitiveness through "agentic AI"—AI systems that can act autonomously and collaboratively.

This reflects a growing recognition in Japan and globally that the next wave of AI value won't come from individual companies deploying isolated AI systems, but from creating AI ecosystems where autonomous agents collaborate across organizational boundaries.

It's a shift from "my AI" to "our AI"—without losing control of "my data."

Challenges and Questions That Remain

Despite the promise, several important questions linger:

Verification and Trust: How do companies verify that the secure gateway is actually preventing data leakage? While Fujitsu's system simulates and monitors communications, building trust in these safeguards will take time.

Standardization: For this to work at scale, there needs to be some level of standardization in how AI agents communicate and negotiate. Will companies rally around a common protocol, or will we see fragmentation?

Performance Trade-offs: Does the privacy-preserving approach sacrifice optimization quality? The 30% cost reduction suggests not, but more data from the upcoming trials will be crucial.

Liability and Governance: If an AI agent makes a bad recommendation that affects multiple companies, who's responsible? The governance frameworks for multi-agent AI systems are still being developed.

Competitive Dynamics: Even with data protection, some companies may resist collaborating with direct competitors. Cultural and strategic factors matter as much as technical capabilities.

What This Means for Supply Chain Professionals

If you're working in supply chain management, procurement, or logistics, here's what to watch:

1. The "data sharing" excuse is disappearing: For years, executives could point to data sensitivity as a valid reason to avoid AI collaboration projects. That argument is becoming harder to sustain.

2. First-mover advantages are real: Companies that adopt multi-agent AI collaboration early will build relationships and optimize their networks ahead of competitors who wait.

3. Internal data quality still matters: Even with secure collaboration, your AI agent is only as good as the data it's trained on. This technology doesn't eliminate the need for clean, well-structured internal data systems.

4. New skills required: Managing collaborations with AI agents from partner companies will require new competencies—a blend of technical understanding, negotiation skills, and strategic thinking.

5. Pilots are coming soon: With Fujitsu targeting commercial availability by the end of fiscal 2026 and other companies likely developing similar capabilities, the window to start small-scale experiments is opening now.

The Bigger Picture: Toward a More Resilient Global Economy

The past few years have dramatically exposed the fragility of global supply chains. From pandemic-related disruptions to geopolitical tensions to climate-related disasters, companies have learned the hard way that isolated optimization isn't enough. True resilience requires coordination.

Fujitsu's multi-AI agent technology represents a potential path forward—one where companies can achieve the benefits of collaboration without sacrificing the protection of competitive information.

It won't solve every supply chain challenge. Labor shortages, infrastructure limitations, and geopolitical risks will persist. But by removing the trust barrier that has prevented companies from fully leveraging AI's optimization capabilities, this technology could mark an inflection point.

The question isn't whether multi-agent AI collaboration will become standard practice. The question is how quickly companies will adopt it—and who will gain the competitive advantage of moving first.

Frequently Asked Questions (FAQ)

Q: How is this different from blockchain for supply chain management?

A: While blockchain provides a shared ledger for transactions, it doesn't solve the AI collaboration problem. Blockchain still requires companies to share data onto the distributed ledger, and it doesn't enable AI systems to learn from each other's patterns without exposing raw data. Fujitsu's approach focuses specifically on allowing AI agents to collaborate and optimize without data sharing, which is a fundamentally different challenge.

Q: Can companies really trust that their data won't be leaked?

A: Fujitsu's system employs multiple security layers: knowledge distillation that shares insights rather than raw data, secure gateways that filter communications, and continuous monitoring to detect malicious queries. However, building complete trust will require independent audits, real-world testing, and time. The upcoming trials with Rohto Pharmaceutical will provide crucial validation.

Q: What happens if one company's AI agent makes a mistake that affects others?

A: This is one of the unresolved governance challenges. Fujitsu's system allows each AI agent to propose solutions based on its understanding of the overall optimization goal, but final decisions and implementations remain with human operators at each company. The negotiation-based approach means no single agent can force actions on others. However, clear contractual frameworks and liability agreements will be essential as these systems mature.

Q: How much does this technology cost to implement?

A: Fujitsu hasn't publicly disclosed pricing yet, as the technology is still in field trials. However, the company plans to offer it through their Uvance business model's Dynamic Supply Chain services by the end of fiscal 2026. Costs will likely vary based on the complexity of the supply chain, number of participating companies, and level of customization required.

Q: Do all companies in a supply chain need to use Fujitsu's system for it to work?

A: Not necessarily. The system is designed to enable collaboration between AI agents from different vendors, not just Fujitsu's own agents. The key requirement is that participating AI systems can communicate through the secure gateway protocol. This vendor-agnostic approach is crucial for widespread adoption, as supply chains typically involve companies with different technology stacks.

Q: What industries beyond supply chain could benefit from this?

A: Any sector where companies need to collaborate but can't share sensitive data could benefit: healthcare (coordinating patient care across providers), financial services (fraud detection and risk assessment), energy (optimizing power grids), smart cities (coordinating transportation and infrastructure), and collaborative research (pharmaceutical development, climate modeling). The core principle—secure AI collaboration without data sharing—has broad applications.

Q: How does this compare to federated learning?

A: Federated learning allows machine learning models to be trained on decentralized data without the data leaving its original location. Fujitsu's approach includes elements of federated learning but goes further by enabling active negotiation and collaboration between AI agents, not just distributed training. It's designed for real-time operational optimization in multi-party environments, not just model training.

Q: What if a competitor deploys this technology first? How big is the competitive disadvantage?

A: Early adopters will likely gain advantages in supply chain efficiency, responsiveness to disruptions, and partner relationships. However, the technology's value increases with network effects—the more companies participate, the better the optimization becomes. This creates an incentive for widespread adoption rather than exclusive use. Companies that wait too long risk being locked out of increasingly efficient collaborative networks.

Q: Can small and medium-sized businesses benefit, or is this only for large enterprises?

A: While the initial trials involve large companies like Rohto Pharmaceutical, the technology's design could scale down. Small suppliers that participate in optimized supply chains with larger partners could benefit from improved demand forecasting and logistics efficiency without needing to build complex AI systems themselves. The SaaS delivery model Fujitsu plans could make this more accessible to SMBs, though pricing and usability will be key factors.

Q: How does this technology handle different languages and regional regulations?

A: Fujitsu hasn't detailed language capabilities yet, but as a global technology company, they're likely building multilingual support. Regional regulations are more complex—different countries have varying data protection laws (GDPR in Europe, CCPA in California, etc.). The secure gateway approach should help with compliance by ensuring data stays within jurisdictional boundaries while still enabling collaboration, but companies will need to evaluate compliance on a case-by-case basis.

Q: What skills do our employees need to work with this technology?

A: Teams will need a combination of technical AI literacy (understanding what AI agents can and can't do), strategic thinking (knowing what to optimize for), negotiation skills (setting parameters for how your AI agent interacts with others), and data governance expertise (ensuring your internal data quality supports effective AI agent performance). Most companies won't need deep AI engineering skills but will need people who can bridge business strategy and AI capabilities.

Q: Is this technology proven, or is it still experimental?

A: As of December 2025, it's in the transition phase from experimental to proven. Initial trials on virtual supply chains showed promising 30% cost reductions, but the real-world trials beginning in January 2026 will provide the critical validation. Fujitsu plans commercial availability by the end of fiscal 2026, suggesting they're confident in the technology's readiness. Early adopters should expect some iteration and refinement, but the core concepts have been demonstrated.