518 VIEWS
Pinterest LinkedIn LINE

On March 6, I appeared at the FINOLAB-hosted Japan Fintech Week-related event “Toward the Implementation of AI Agent Payments: From Governance to Authentication and Authorization”. I was supposed to have a back-and-forth with Mr. Shibata, but despite the script that had so kindly been prepared for us, I completely went off script. Building on the earlier presentations by Dentsu Soken, the Japan Research Institute, and the AI Governance Association, I improvised almost the entire session and turned it into a discussion involving the audience as well.

As a result, I do not have anything left on hand that I can simply copy and paste, but I think I said something along the following lines.

  • Giving an AI agent your own key, or a token intended for another client, as-is is semantically what is known as “impersonation,” and must not be done. It must instead be handled as “delegation.”
  • To do that, an AI agent must be assigned an independent identity, just like an employee.
  • However, the number of agents is expected to reach as many as 100 per employee, which will overwhelm the systems currently in use.
  • It is currently said that an IGA solution becomes necessary once an organization has more than 1000 employees. In the age of AI agents, however, even a small business with 10 human employees could cross that threshold.
  • Moreover, even if we say that an identity should be assigned, the unit to treat as the Entity requires careful consideration. Does changing the model make it a different person? If different models inherit or share the same memory, are they the same Entity?
  • All of this must be Audit by Default: information about what was done and why must be gathered in real time and delivered to the human who delegated the work.
  • However, because humans can no longer review all of it themselves, we also need agents that monitor and triage it, escalate matters to humans when necessary, and in some cases activate a Kill Switch.

And so on.

It seems the session was also recorded, so I expect a report will be published in due course.

The following is from the event announcement page.

At this seminar, held during Japan FinTech Week, we will address issues including governance, privacy, payment use cases, and authentication and authorization in order to introduce AI agents into financial operations such as payments. We plan to discuss the path toward future implementation, and warmly invite you to join us.

◆Date and time: March 6, 2026 (Fri), 14:00-17:00 (doors open at 13:45)
◆Venue: FINOLAB Event Space
, 1-6-1 Otemachi, Chiyoda-ku, Tokyo, Otemachi Building 4F
(The Otemachi Building runs east to west; the entrance is at its eastern end, on the Tokyo Station side.)  
◆Language: Japanese (English-language materials will also be provided)
◆Admission: Free
◆Organizer: Dentsu Soken Inc.
◆Supported by: FINOLAB Inc.; Privacy by Design Lab
◆Agenda:
14:00₋14:05 Opening remarks (Shinsuke Sano, Division General Manager, Dentsu Soken Inc.)
14:05-14:30 “AI Agent Payments with Real-World Operations in Mind” (Kohei Ichihara, Deputy Director and Senior Researcher, Advanced Technology Laboratory, The Japan Research Institute, Limited; concurrently, Digital Strategy Department, Sumitomo Mitsui Banking Corporation)
14:30-15:00 “Authentication and Authorization Critical to AI Payments and Reservations + Demo” (Tetsuaki Fukushima, Chief Producer, Dentsu Soken Inc.
              / Yasuo Higa, Strategy and Solutions Architect → Kazuhiro Toda, Group Manager, Digital Trust Group, Open Innovation Laboratory)
15:00-15:25 “AI Governance” (Hiroaki Sakuma, Executive Director and Secretary General, AI Governance Association)
15:25-15:55 “Identity Governance” (Natsuhiko Sakimura, Digital Identity Advisor, PwC Japan Group / Takehisa Shibata, Director, PwC Consulting LLC)
15:55-16:00 Closing remarks (Kuniharu Sakai, Unit Head, Dentsu Soken Inc.)
16:00-17:00 Networking

Related posts