[TOC Artificial Intelligence Initiatives Meeting] 2025-08-29

[riaankleinhans]

Host:
Note Taker:

Recurring Items

• LF Antitrust Policy
• TAG Folder
• TOC SubProjects Folder
• Welcome new meeting attendees

Agenda

[joshhalley]

@raravena80

Session scheduled for this week:
Title: “Rethinking Zero Trust Models for Agentic AI”
Presenter: Marcelo Yannuzzi, Ph.D. – Cisco
Duration: 35 mins

++++++++++++++++++++++++++++++++++++++++++++++++++++++
Abstract:
The explosive growth of AI Agents and MCP Servers is making evident the various technology gaps that exist today to identify these entities, establish trust, define policies, enforce access controls, and securely delegate the execution of tasks to autonomous entities that operate at machine speed.
This is not automation as we knew it. It’s a different ball game that demands structural changes to current Zero Trust models, including (re)authentication and MFA, delegated authorization, access controls, micro-segmentation, and more.
In this session, we will share how we are addressing some of these problems, our initial open source contributions through AGNTCY (https://agntcy.org/), and our efforts to bring collective thinking around these new challenges.
We will cover how to generate, assign, and provide verifiable identities to AI Agents and MCP Servers, how to build trust even before attempting to connect to these entities, our current thinking about more granular access controls, along with delegated authorization with and without a human in the loop.
++++++++++++++++++++++++++++++++++++++++++++++++++++++

[joshhalley]

Hosts:

@raravena80
@joshhalley
@zanetworker

Attendees:
Riaan Kleinhans: Linux Foundation
Nina Polshankova: solo.io (agent gateway integrations)
Deep Patel: Cisco (datacenter security)
Marcello Yannuzzi: Cisco (Outshift)
Yujia Lin: pending update
Amandeep Singh: pending update
Nimisha Mehta: pending update
Satarupa Deb: pending update
Mahendran Selvakumar: pending update
Raghu Shankar: pending update
Paras Patel: pending update
Ronald Petty: RX-M
Andrey Velichkevich: Apple
Sudhanshu Prajapati

Events:
OSS Summit / AI_DEV Amsterdam: Aug 28th - 29th (concluding today)
KCD Cloud Native Day SF: Sep 9th - 10th (Ron shared an update with the team)
Kubecon North America: Nov 10th - 13th

Standing items
WG / Initiative updates
Issues in repo: CNCF CN AI Working Group Project (view)

Security Paper [DP] AI Sec Whitepaper link

• Review of paper took place, request to condense the content to ~10 pages per section, Deep is working on reductions at the moment, should tie up this weekend

Kubernetes Agentic AI Standards Checklist

• Vote concluded with approval
• Final actions are being worked through before taking the item forward, follow up planned next week

Scheduling Paper Updates [CM] [PUBLIC] Scheduling for Cloud Native AI Whitepaper

• ACTION: Looking for somebody to take leadership of this/these paper(s) to drive to closure
• Kai Scheduling team have indicated that they want to contribute, however a lead is still required for this item.
• Ron raised the topic of credits and proofs from the data in this paper (team to follow up)

Sustainability: [WHITEPAPER] Sustainable AI in Cloud-Native Environments

• ACTION: Looking for somebody to take leadership of this/these paper(s) to drive to closure
• Already split into two papers ([Initiative]: Cloud Native AI Sustainability Papers #1675)
• Needs to be split into two initiatives

Issue (#1746) raised by Vincent Caldeira, proposing the following topics to be discussed:
Protocol interoperability
Agentic Gateway / Data Plane
Runtime abstraction
Fault Tolerance
Observability & Policy Management

MCP Update on the K8s context Standardize MCP Servers and Tool Ecosystem Across Cloud-Native Categories tag-runtime#212

OTEL has a semantic convention WG doing a lot of work on AI at the moement - Ricardo Aravena/Josh Halley OpenTelemetry LLM Semantic Convention WG

• New update shared on AI Agent SPANS in otel: AI Agent Span semantics, a lot of work is happening in this space at the moment
• One noteable disappointment is in the adjacent gitops space, https://gitlab.com/gitlab-org/gitlab/-/issues/338943 while not directly linked to AI, end-to-end pipeline execution flows lack will not be supported. In contrast, jenkins on the other hand, are enriching their tooling with OTEL and correct semconvs.

Topic of MCP Whitepaper was raised by Satarupa Deb - Josh & Ricardo to align on topic and reach out to Satarupa on recommended next steps.

Presentation: “Rethinking Zero Trust Models for Agentic AI”
Presenter: Marcelo Yannuzzi, Ph.D. – Cisco
Duration: 35 mins

++++++++++++++++++++++++++++++++++++++++++++++++++++++
Abstract:
The explosive growth of AI Agents and MCP Servers is making evident the various technology gaps that exist today to identify these entities, establish trust, define policies, enforce access controls, and securely delegate the execution of tasks to autonomous entities that operate at machine speed.
This is not automation as we knew it. It’s a different ball game that demands structural changes to current Zero Trust models, including (re)authentication and MFA, delegated authorization, access controls, micro-segmentation, and more.
In this session, we will share how we are addressing some of these problems, our initial open source contributions through AGNTCY (https://agntcy.org/), and our efforts to bring collective thinking around these new challenges.
We will cover how to generate, assign, and provide verifiable identities to AI Agents and MCP Servers, how to build trust even before attempting to connect to these entities, our current thinking about more granular access controls, along with delegated authorization with and without a human in the loop.
++++++++++++++++++++++++++++++++++++++++++++++++++++++

[sudhanshu456]

Hi @joshhalley, could you add me to the attendees list? I don't see my name. I was having some issues with zoom meeting, so I joined approximately 30 minutes late.