Experience the difference. Meet your peers, talk with analysts, build your network and turn insights into action.
Akira Otsuka
Graduate School of Information Security, Institute of Information Security (IISEC) / Prof. Dr.
Speaker Topics
- Using AI in Cybersecurity
Speaker Bio
Akira Otsuka received the B.E. and the M.E. degree from Osaka University in 1989 and 1991, respectively, and received the Ph.D. degree from the University of Tokyo in 2002. From 2005, he was with National Institute of Advanced Industrial Science and Technology (AIST), where he was a Leader of Research Security Fundamentals during 2006 to 2010. From 2007 to 2014, he was a Visiting Professor at the Research and Development Initiative, Chuo University. He was a Visiting Researcher at he Institute for Monetary and Economic Studies (IMES), Bank of Japan (2020-2021). Since 2017, he has been a Professor with the Graduate School of Information Security, Institute of Information Security in Yokohama, Japan. He is Senior Member of IEICE, IPSJ, and Member of JSAI, IACR, and IFCA. He serves as a leader of AI Safety and Security Research Group (SIG SEC) at JSAI, and a seering committee chair of the JNSA Industry–Academia Collaboration Council for Cybersecurity. His research interest includes the interdiciplinary research area of Cryptology and Machine Learning, particularly in Cryptologic approaches to AI security.
Show More
Show Less
Sessions
Wednesday, 22 July, 2026 05:45 PM - 06:30 PM JST
Guest Keynote: Agentic AI Security
Akira Otsuka,
Graduate School of Information Security, Institute of Information Security (IISEC) / Prof. Dr.
The focus of security research on agentic AI is rapidly shifting from the safety of standalone models to the security of entire systems that integrate planning, tool use, memory, and multi-agent coordination. It is no longer sufficient for an LLM simply to refrain from generating harmful text. What matters increasingly is whether an agent can be manipulated—accidentally or deliberately—while formulating plans, invoking tools, consulting memory, and adaptively revising its goals in response to external inputs. Research from Stanford University, exemplified by Cybench, demonstrates that state-of-the-art LLMs can solve difficult Capture the Flag (CTF) challenges used in hacker competitions with striking speed, underscoring the inseparability of capability evaluation and misuse-risk assessment. Meanwhile, OpenAgentSafety, developed by researchers at Carnegie Mellon University and the Allen Institute for AI, introduces more than 350 multi-turn tasks in realistic tool-use environments involving browsers, code execution, file systems, bash, and messaging. Their results show that current models still exhibit unsafe or inappropriate behaviors at a high rate, revealing the limitations of existing safeguards and the need for more robust defensive techniques. This talk will examine the security threats posed by agentic AI through concrete case studies and discuss several emerging approaches to mitigation, drawing on recent research in the field.
...
Show More
Show Less
Thursday, 23 July, 2026 02:15 PM - 03:00 PM JST
CISO Circle Roundtable: Agentic AI: How Should We Think About It and How Should We Utilize It?
Akira Otsuka,
Graduate School of Information Security, Institute of Information Security (IISEC) / Prof. Dr.
In this roundtable, based on research trends and examples from the lecture, we will discuss how organizations should position agentic AI and under what conditions it can be safely used. Agentic AI has powerful capabilities and it is essential to understand the potential risks, limitations of current models and countermeasures. Moving forward with utilization while correctly recognizing the risks and challenges is an important step toward realizing the value of agentic AI in the future.
...
Show More
Show Less