RAYVEN CEO Yoshihiro Suzuyama Wins Grand Prize at METI-JETRO Sponsored "Shido Next Innovator" DEMO DAY
Yoshihiro Suzuyama, CEO of RAYVEN Inc., which develops AI agent governance infrastructure, has won the Grand Prize at the 11th "Shido Next Innovator" DEMO DAY, sponsored by the Ministry of Economy, Trade and Industry (METI) and JETRO. His proposal for a "mechanism for companies to control the authority and actions of AI agents" received the highest evaluation.
📋 Article Processing Timeline
- 📰 Published: March 31, 2026 at 18:10
- 🔍 Collected: April 1, 2026 at 13:39 (19h 29m after Published)
- 🤖 AI Analyzed: April 22, 2026 at 00:39 (491h 0m after Collected)
RAYVEN Inc. (Headquarters: Kita-ku, Osaka), which develops AI agent governance infrastructure, announced that its CEO, Yoshihiro Suzuyama, received the Grand Prize at the 11th "Shido Next Innovator" DEMO DAY, an entrepreneur development program sponsored by the Ministry of Economy, Trade and Industry (METI) and JETRO.
In this program, where 100 selected entrepreneurs from across Japan refine their businesses over approximately six months, and 10 individuals dispatched to Silicon Valley present their final pitches, Suzuyama's proposal for a "mechanism for companies to control the authority and actions of AI agents" garnered the highest evaluation.
## ■ AI Agent Adoption Halted by 'Security Review'
In 2026, as the business application of AI agents fully takes off, there has been a rapid increase in cases where adoption is stalled at the 'security review' stage. This is not limited to industries with strict security requirements such as finance, manufacturing, and public sectors. Even in general companies that have begun to incorporate AI into their operations, introduction plans are being halted as they cannot answer questions like 'To what extent should AI be granted access?' and 'Who is responsible for an accident if one occurs?'
The Model Context Protocol (MCP), a standard protocol used when AI agents connect to internal systems and external tools, has been pointed out to have vulnerabilities that directly lead to management risks, regardless of industry.
* **Execution of unintended operations (Tool Poisoning)**
Risk of AI executing customer data deletion or fraudulent processing due to malicious tool definitions.
* **Access to information beyond authority (Privilege Escalation)**
Risk of AI accessing confidential information it is not authorized to view, leading to information leakage.
* **Absence of action logs**
Risk of inability to identify the cause or respond to audits in the event of an incident due to no record of 'who instructed what.'
According to an Equixly survey in March 2025, command injection vulnerabilities were confirmed in 43% of major OSS MCP servers. OWASP also highlights security challenges with MCP servers, indicating that establishing governance is becoming an unavoidable management challenge for all companies adopting AI.
## ■ Yoshihiro Suzuyama's 'AI Employee ID' Concept
Human employees are issued ID cards upon joining a company, have restricted access to systems, and their actions are recorded. In contrast, AI agents lack this system.
Suzuyama is leading the development of a governance platform that provides AI agents with the same level of 'authentication, authorization, and auditing' as human employees. While traditional AI tools and API management control 'entry' and 'exit,' RAYVEN's technology intervenes directly in MCP communication to verify, control, and record in real-time which tools AI calls, which data it accesses, and what it attempts to execute.
## 4 Technical Features
1. **Enforcement of Authority**
In this program, where 100 selected entrepreneurs from across Japan refine their businesses over approximately six months, and 10 individuals dispatched to Silicon Valley present their final pitches, Suzuyama's proposal for a "mechanism for companies to control the authority and actions of AI agents" garnered the highest evaluation.
## ■ AI Agent Adoption Halted by 'Security Review'
In 2026, as the business application of AI agents fully takes off, there has been a rapid increase in cases where adoption is stalled at the 'security review' stage. This is not limited to industries with strict security requirements such as finance, manufacturing, and public sectors. Even in general companies that have begun to incorporate AI into their operations, introduction plans are being halted as they cannot answer questions like 'To what extent should AI be granted access?' and 'Who is responsible for an accident if one occurs?'
The Model Context Protocol (MCP), a standard protocol used when AI agents connect to internal systems and external tools, has been pointed out to have vulnerabilities that directly lead to management risks, regardless of industry.
* **Execution of unintended operations (Tool Poisoning)**
Risk of AI executing customer data deletion or fraudulent processing due to malicious tool definitions.
* **Access to information beyond authority (Privilege Escalation)**
Risk of AI accessing confidential information it is not authorized to view, leading to information leakage.
* **Absence of action logs**
Risk of inability to identify the cause or respond to audits in the event of an incident due to no record of 'who instructed what.'
According to an Equixly survey in March 2025, command injection vulnerabilities were confirmed in 43% of major OSS MCP servers. OWASP also highlights security challenges with MCP servers, indicating that establishing governance is becoming an unavoidable management challenge for all companies adopting AI.
## ■ Yoshihiro Suzuyama's 'AI Employee ID' Concept
Human employees are issued ID cards upon joining a company, have restricted access to systems, and their actions are recorded. In contrast, AI agents lack this system.
Suzuyama is leading the development of a governance platform that provides AI agents with the same level of 'authentication, authorization, and auditing' as human employees. While traditional AI tools and API management control 'entry' and 'exit,' RAYVEN's technology intervenes directly in MCP communication to verify, control, and record in real-time which tools AI calls, which data it accesses, and what it attempts to execute.
## 4 Technical Features
1. **Enforcement of Authority**