TLDR: SoftBank has developed a prototype “Agent Firewall” technology designed to enable secure communication and collaboration between AI agents from different organizations, addressing challenges related to data governance and compliance. The system, which incorporates dynamic governance, authentication, policy enforcement, and real-time auditing, is expected to be commercialized around fiscal year 2026.
SoftBank Corp. has announced the development of a groundbreaking “Agent Firewall” technology, a prototype designed to facilitate secure and flexible collaboration among artificial intelligence agents across various organizations. This innovation directly addresses critical challenges in multi-AI agent environments, particularly concerning data governance, access control, and regulatory compliance when data exchanges occur across different corporate or national boundaries.
The telecommunications giant plans to commercialize this system around fiscal year 2026, following extensive internal testing and collaborations with partners. The “Agent Firewall” is poised to overcome anticipated obstacles in environments where enterprises operate under diverse data handling rules, management standards, and data location requirements, which could otherwise impede inter-organizational data collaboration.
Key features of SoftBank’s “Agent Firewall” include:
Dynamic Governance Control: The system automatically adjusts the intensity of inspections based on communication risk levels and required security standards.
Authentication and Authorization: It provides robust functions to ensure that communication is permitted only between trusted AI agents.
Contract and Policy-Based Controls: The firewall automatically detects violations in accordance with contractual conditions and organizational policies, thereby preventing data loss incidents.
Monitoring, Auditing, and Log Management: Communications are recorded in tamper-resistant ledgers with real-time monitoring capabilities.
SoftBank emphasizes that this technology complements existing protocols, such as the Agent-to-Agent Protocol (A2A) and Agent Communication Protocol (ACP), ensuring consistent application of security policies regardless of the protocol type. The system operates on “zero-trust” principles, moving beyond traditional network perimeter security approaches by checking only the necessary data, which enhances efficiency and scalability.
Also Read:
- Fujitsu Unveils Multi-AI Agent Security System to Combat Evolving Cyber Threats
- AI Agents Revolutionize Cybersecurity: A New Era of Human-Machine Collaboration Emerges
This development is part of SoftBank’s broader strategy to accelerate secure AI teamwork on new data platforms and contribute to building trust and interoperability within complex, distributed AI systems. The company has already filed patent applications for the technology and aims to integrate it with sovereign cloud and sovereign AI systems, catering to corporations, municipalities, and research institutions handling sensitive data.
