CNCF [Cloud Native Computing Foundation]

Sponsored Keynote | Midships Global: You Don't Need a New Identity Platform for the AI Era - Y. Lin

Duration: 4:26 691 transcript words EN summary

Video summary

The rapid emergence of AI agents marks a significant shift in enterprise architecture, moving beyond traditional user-interface interactions to autonomous systems capable of reasoning, planning, and executing tasks independently. This evolution introduces a new category of actors that necessitates addressing critical identity questions regarding authentication, authorization, and the enforcement of human-in-the-loop controls. Despite these novel challenges, the core requirements remain rooted in established domains such as access management, audit trails, and delegation rather than requiring entirely new paradigms. A key argument presented is that the industry's push to replace existing infrastructure with purpose-built AI identity platforms is unnecessary and potentially misguided. When a major UK bank suggested that the introduction of agents required a completely new identity platform due to perceived risks, this view was challenged by evaluating whether fundamental identity primitives had actually changed. The conclusion reached was that they have not; instead of reinventing the wheel, organizations can leverage existing solutions to handle these complexities effectively while maintaining centralized control over context-based authorization and just-in-time least privilege access. To demonstrate a practical approach, a tier-one bank in the Philippines successfully implemented agent identities using Keycloak alongside its human identity management system. This solution utilized delegation and token exchange capabilities to securely chain interactions between humans, agents, and tools, while integrating with SPIFFE and SPIRE to eliminate hardcoded credentials. Furthermore, the implementation enforced strict policies for high-risk actions requiring human approval and utilized DPoP under FAPI 2.0 to validate human attestations, proving that robust agentic identity can be achieved by carefully assembling existing open-source foundations rather than adopting proprietary AI-specific platforms. Ultimately, the experience shared underscores that enterprises do not need a new identity platform to navigate the AI era but rather need to thoughtfully integrate current capabilities to support autonomous agents. By leveraging the flexibility of open-source tools like Keycloak, organizations can piece together a comprehensive security framework that addresses the unique needs of agentic workflows without starting from scratch. This approach validates the notion that the foundations of identity management remain solid and sufficient, allowing businesses to adapt to dynamic AI environments efficiently and securely.

Read the full video transcript

I'm E-Sam from E-Ships. So, I have been working on IM solutions for nearly a decade and we work with some of the biggest banks in the world on their IM platform. So, I'm here today to talk to you on my experience in achieving agent identity with Keycloak. So, I believe all of us recognize that we are moving to a AI agent era where enterprises are actually deploying systems that lavish the power of LLMs to behave in a very dynamic and autonomous ways that we never expect from traditional software applications. So, for the longest time we expect a user to interact with applications UI and that application is integrated with various APIs and invoke those API based on the human facilitation. Today, the human interact with a human with a AI agent and that AI agent can reason, create plans, execute their plans to achieve what the human wants. And that actually creates a new category of actors in our architecture. And this new category of actors will flash our mind with a lot of identity access questions. For example, how can the agents be authorized and authenticated? How do we know who the agent is acting on behalf of? A human or agent? How far the authorization is moving down the chain? How do we enforce human in the loop? So, while all these problems can be very overwhelming, but they are not new. In the end, they are still within the domain of authentication, authorization, audit trails, delegations. And we have been in the identity space, we have been solving these problems within different context for a very long time. So, what caught my attention is when one of our client, which is a tier one bank in the UK, told us that their advice by one of the big four uh that because of this new category of actors, the new consideration, the new risk, you need a new identity platform. As someone who has been working in this space every day for the past 10 years, when I received this news, I questioned myself, did the identity primitives really changed? And after evaluation, our conclusion is no, it's not. And our client, which is the UK bank, also agreed with us. We do not want to reinvent identity. So, in contrast, for another tier one bank, this is uh in the Philippines, we actually achieved agent identities uh using Keycloak uh in that bank, and we're also using uh Keycloak in that bank for human identities. So, this is what we did. So, first, we use uh delegation and token exchange uh capabilities within Keycloak to actually chain up humans to agents, agent to agents, agents to tools and APIs. We use fine-grained authorizations and policies within Keycloak to actually achieved uh context-based authorization with centralized control. We also enforced kind of just-in-time least privilege. We enforced for all the high-risk action to require all bank human approvals. And to give you some details uh to complete the picture, we also used uh DPoP under FAPI 2.0 to actually to bring out human attestations. Uh we integrated Keycloak uh with uh uh SPIFFE and SPIRE so that the agents do not need to be hardcoded with any form of credentials. And we also use events to actually con- construct uh context-rich audit trails as the agent is going through identity and authorization processes with Keycloak. So, from our experience, I would say that you do not need a purpose-built AI platform uh to in this new era. What we have done in Midships, what we did was all we are just using existing concepts and foundations and piece them together in a very carefully thought through manner to achieve the agentic identity. And given the Keycloak nature, since it's open source, since Meshits can do it, you can also do it. So, if you have any questions or any doubts regarding how we have achieved agentic identity using Keycloak, I will be around and feel free to talk to me. And that will be the end of my Thank you for your time.