Fujitsu develops world’s first multi-AI agent security technology to protect against vulnerabilities and new threats : Fujitsu Global

Revolutionizing Cybersecurity: The Rise of Collaborative AI Agents

Fujitsu Limited

Kawasaki, Japan, December 12, 2024 – Fujitsu is stepping up its cybersecurity game with a groundbreaking multi-AI agent security system designed to stay ahead of increasingly complex cyber threats. This innovative technology, part of Fujitsu’s AI service platform, Kozuchi, not only safeguards organizations from potential cyberattacks but also actively simulates attack scenarios, enhancing overall security resilience.

At the heart of this system are three key components: the Security AI agent, collaborative multi-AI technology, and enhancements specifically for generative AI security. This trifecta allows organizations to proactively manage vulnerabilities and anticipate emerging threats in real time.

Fujitsu is set to kick off field trials for the generative AI security component in collaboration with Cohere Inc. starting December 2024. Additionally, the company has plans to release part of its innovative multi-agent collaboration technology as open-source software on the OpenHands platform in January 2025, with a comprehensive global rollout aiming for March 2025.

This multi-AI agent security technology enables IT professionals, regardless of their security expertise, to design applications that fortify their organizations against cyber threats. It paves the way for secure incorporation of generative AI into enterprise infrastructures, allowing companies to harness the advantages of this technology while also managing its inherent risks.

Fujitsu’s initiative is all about fostering a trustworthy digital environment conducive to sustainable innovation, embodying their commitment to aligning technological advances with societal well-being.

Overview of the multi-AI agent security technology

Digging Deeper into the Technology

1. Security AI Agent

This new technology integrates three foundational AI agents tailored for proactive security:

• Attack AI agent: Simulates various attack techniques to propose potential attack scenarios against specific systems.
• Defense AI agent: Develops countermeasures informed by the organization’s unique risk profile.
• Test AI agent: Crafts a virtual testing environment from the real system to gauge the efficacy of defenses.

The collaboration between these agents occurs in a ‘cyber twin’—a virtual replica where they can interchangeably simulate attacks and defenses. This co-creative model leverages diverse skills among agents to tackle complex cybersecurity challenges head-on.

Collaborative synergy among the three security AI agents

2. Multi-AI Agent Collaboration

One of the standout features is the seamless collaboration facilitated between the AI agents, which extends beyond mere security applications. Recognizing the importance of data confidentiality and operating within established policies, Fujitsu’s technology incorporates a secure agent gateway that autonomously governs collaboration protocols. This ensures the AI agents can operate cooperatively without compromising sensitive data.

3. Generative AI Security Enhancement

In light of new threats targeting generative AI, Fujitsu has partnered with Ben-Gurion University to craft a powerful LLM vulnerability scanner. This tool is adept at detecting over 3,500 known vulnerabilities while employing adaptive prompting technology for tailored attack evaluations. Complementing it are LLM guardrails that automatically shield against malicious input, ensuring secure operations in enterprise settings.

Innovative generative AI security enhancement technology

Expert Opinions

Carnegie Mellon University’s Associate Professor Graham Neubig:

“The rapid advancement of AI agents is reshaping the landscape. OpenHands paves the way for diverse applications and services that, when combined with Fujitsu’s technology, will enhance cross-organizational functionality remarkably.”

Professor Yuval Elovici from Ben-Gurion University:

“Fujitsu’s groundbreaking cyber twin technology enables realistic testing of security policies without any real-world impact, effectively addressing new vulnerabilities with agility.”

Seraphina Goldfarb-Tarrant, Head of Safety at Cohere Inc:

“As enterprises embrace AI, secure systems are paramount. We’re thrilled that our collaboration with Fujitsu will empower Japanese firms to utilize our top-notch security and multilingual models.”

Future Aspirations

Looking ahead, Fujitsu envisions broadening its AI capabilities to automate various security tasks further, including secure design processes and incident response measures. The focus will also be on refining the generative AI security enhancements to safeguard against vulnerabilities specific to retrieval-augmented generation applications.

Fujitsu’s Commitment to Sustainable Development

Press Contacts

All trademarks and registered trademarks of companies and products mentioned herein belong to their respective owners. The information provided is accurate as of publication and may change without notice.

Date: 12 December 2024
City: Kawasaki, Japan
Company: Fujitsu Limited