On March 6th, I had the honour to speak at a Japan Fintech Week event hosted by FINOLAB titled "Towards the implementation of AI agent payment: from governance to authentication and authorization". It was intended to be a dialogue with Shibata-san, and despite he had provided me with a script, I completely ignored it and made it very dynamic involving the floor, developing on the themes of the previous sessions by Dentsu Institute of Human Studies, Japan Research Institute, and AI Governance Association.
Because of this, I don't have anything on hand that I can copy and paste, but I think I probably said something like this:
- Handing over your own key or a token for another client to an AI agent is what is known as "impersonation" and should not be done. Instead, we should use "delegation."
- To do this, AI agents must be assigned independent identities, much like employees.
- However, it is thought that the number of agents per employee will reach 100, and the current system will no longer be viable.
- Currently, it is said that if a company has more than 1000 employees, it will need to implement an IGA solution, but in the age of AI agents, even small companies with 10 people could fall prey to this.
- Also, when assigning an identity, it is necessary to consider which unit to define as an entity. If the model changes, is it a different person? If different models inherit/share memory, is it the same entity?
- These must be audited by default, and information about what was done and why must be collected in real time by the person who commissioned the work.
- However, since these are no longer visible to the human eye, we will need agents to monitor them, triage them, escalate them to humans if necessary, and in some cases activate the kill switch.
etc.
It seems that they also recorded it, so I think a report will be released soon.
Below is from the announcement page.
This seminar will be held during Japan FinTech Week. We will discuss issues such as governance, privacy, payment examples, authentication and authorization in order to introduce AI agents into financial operations such as payments, and will also discuss future implementations. Please join us!
◆Date and Time : March 6, 2026 (Friday) 14:00-17:00 (doors open 13:45)
◆Venue : FINOLABEvent Space
Otemachi Building 1F, 6-1-4 Otemachi, Chiyoda-ku, Tokyo
*Otemachi Building is a long building stretching from east to west, with the entrance at the eastern end, on the Tokyo Station side.
◆Language: Japanese (English materials also available)
◆Participation fee:free
◆Sponsorship : Dentsu Research Institute, Inc.
◆Sponsorship : FINOLAB Co., Ltd. General incorporated association Privacy by Design Lab
◆Agenda :
14:00-14:05 Opening (Shinsuke Sano, General Manager, Dentsu Institute of Human Studies)
14:05-14:30 "AI Agent Payment with a Practical Perspective" (Japan Research Institute, Limited Deputy Director of the Advanced Technology Lab, Senior Researcher,Sumitomo Mitsui Banking Corporation Kohei Ichihara, Digital Strategy Department)
14:30-15:00 "Authentication and Authorization: Key to AI Payment and Reservations + Demo" (Tetsuaki Fukushima, Chief Producer, Dentsu Institute of Human Studies)
/ Yasuo Higa, Strategy and Solution Architect → Open Innovation Lab Digital Trust Group Group Manager Toda Kazuhiro)
15:00-15:25 "AI Governance" (AI Governance Association Executive Director and Secretary General: Hiroaki Sakuma
15:25-15:55 "ID Governance" (PwC Japan Group Digital Identity Advisor Natsuhiko Sakimura/PwC Consulting LLC Director: Takehisa Shibata)
15:55-16:00 Closing remarks (Kuniharu Sakai, Unit Leader, Dentsu Institute of Human Studies)
16:00-17:00 Networking