How should security change when software agents can act on behalf of people? In this conversation from IBM Think 2026, Sarbjeet Johal speaks with IBM’s Suja Viswesan about securing agentic AI at runtime.
Identity, access, and accountability for agents
The discussion examines why controls designed for human users may need to evolve for agents. Topics include traceability, accountability, agentic identity and access management, and the risks of accumulating access or escalating privileges.
Security throughout an agent’s operation
Johal and Viswesan explore the distinction between runtime security and design-time security, as well as the relationship between human and agent privileges. They also discuss retaining context while managing permissions, the implications of composite applications, and security across products and vendors.
Explore the conversation
- 00:54 — The changing nature of agentic security
- 02:36 — Traceability and accountability
- 04:26 — Agentic identity and access management
- 07:25 — Retaining context and managing privileges
- 10:42 — Runtime and design-time security
- 12:30 — Human and agent privileges
- 16:35 — Security across products and vendors
Video published on YouTube on June 25, 2026. Watch the original conversation on YouTube.