SoftBank develops “Agent Firewall” for safe AI collaboration

SoftBank Corp. has revealed a prototype called “Agent Firewall.” This security framework helps AI agents work together safely and flexibly, even across different organizations. AI ecosystems are moving toward multi-agent collaboration to improve society. However, challenges have arisen. These include data governance, access control, and regulatory compliance. This is especially true when data crosses corporate or national borders.

The Agent Firewall tackles these issues. It lets AI agents communicate in a controlled and monitored way. This also keeps data sovereignty intact. Key features include dynamic governance that adjusts inspection levels based on risk, agent-level authentication and authorization, automatic enforcement of contract and policy conditions, and real-time audit logging using tamper-resistant ledgers. The system supports protocols like A2A and ACP. This ensures security stays consistent, no matter the protocol type.

こちらもお読みください： Takenaka introduces DeCYFIR for real-time Cyber Threats

The firewall uses a “zero trust” model. It checks only the data it needs. This approach does not depend on old network boundaries. As a result, it improves efficiency and scalability. This innovation enables ソフトバンク to ease secure AI teamwork on new data platforms. They aim to deploy it by fiscal 2026. The development shows a wider industry effort. It aims to build trust and interoperability in complex, distributed AI systems.