Talks, panels & podcasts
Conference talks, panels, podcasts and webinars, mostly about authorization, AI agents and open standards. Where there's a recording, there's a link to watch or listen.
18 cities in 8 countries. 72 of the 99 sessions below have a recording you can watch or listen to.
Authorization is the last improvised layer of identity. Why AI agents (which need access at runtime and can't be pre-provisioned) and AI-written code are forcing authorization out of the app, and why AuthZEN makes externalizing it an architecture decision instead of a vendor bet.
Authorization used to live in one codebase. Now every layer makes its own decision, and the failures show up as outages, retry storms and "can't reproduce" tickets. The fix: one policy decision point, coordinated with OpenID AuthZEN.
What starts as a check on the user's email domain turns into a web of if/else statements. The stages every company goes through with permissions, and a GitOps approach to scaling policy.
Querying terabytes isn't cool any more. Interactive analytics against a 1 petabyte dataset in BigQuery, live on stage, and why it was cheaper and easier than it sounds. Still my favourite talk.
99 of 99
Why authorization failures are turning into reliability incidents, and how one policy decision point plus OpenID AuthZEN stops each layer of the stack making up its own answer.
With Mike Shema
A reference architecture for agent access: sandbox the agent, route every tool call through a gateway, decide at a policy layer, log the lot. Live demo of Cerbos and Tailscale Aperture deciding on each tool call.
With Simon Moffatt
Authorization is the last improvised layer of identity. Why AI agents (which need access at runtime and can't be pre-provisioned) and AI-written code are forcing authorization out of the app, and why AuthZEN makes externalizing it an architecture decision instead of a vendor bet.
What happens when an AI agent needs more privileges than the person who launched it, and how to decide, per tool call, whether it gets them.
With Vatsal Gupta (Apple)
Access reviews assume access is something people have. In modern systems it's something that gets calculated, so governance has to move from reviewing entitlements to governing policy intent, enforcement and the decisions that actually happened.
With Atul Tulshibagwale (CrowdStrike), Mark Berg (Axiomatics)
A 2-hour masterclass on the AuthZEN Authorization API: the spec's internals, what implementers need to know, certification, and authorizing AI agents with COAZ.
With Atul Tulshibagwale (CrowdStrike), Mark Berg (Axiomatics)
Authentication is largely solved and authorization isn't. An update from the AuthZEN co-chairs on how Shared Signals, AuthZEN and Transaction Tokens fit together.
Where the authorization standards actually stand (AuthZEN, OAuth RAR, CAEP and RISC), and which bits you can use today versus which are still theory.
With Reiner Mertens (KuppingerCole, moderator)
Agents are workloads that spin up, get permissions, do something and die. You can't wait 24 hours for an IGA process, so the building blocks (OAuth, AuthZEN, Shared Signals) have to make access reactive and signal-driven.
The Authorization API 1.0 a few months after it went final, and what the working group is tackling next.
Authorization used to live in one codebase. Now every layer makes its own decision, and the failures show up as outages, retry storms and "can't reproduce" tickets. The fix: one policy decision point, coordinated with OpenID AuthZEN.
With Atul Tulshibagwale (CrowdStrike)
Per-tool-call authorization for AI agents, with the AuthZEN MCP profile as the wire format.
Where the AuthZEN specs had got to the week before IIW. The player jumps straight to my section.
With Atul Tulshibagwale, Mike Kiser (SailPoint), Dick Hardt (Hellō)
AuthZEN, Shared Signals, SCIM events and IPSIE, and what they add up to for enterprise identity. My segment starts 3 hours 43 minutes into the recording (the player jumps straight there).
James Reason's Swiss cheese model from aviation safety, mapped onto the six layers of zero trust, with authorization as the layer that holds the rest together.
Where the AuthZEN spec and working group had got to, then a live demo of identity providers calling out to policy engines through AuthZEN.
The Swiss cheese talk, reworked for a room of CISOs and security leads. The AV died halfway through, so the second half turned into 15 minutes of Q&A on AI agents and workload identity (which honestly went better).
With Omri Gazitt (Aserto), David Brossard (Axiomatics)
With Mike Schwartz
Recorded live at KubeCon North America.
With Pieter Kasselman, Christian Posta, Brett Caley, Andrew Block
With Luca Maraschi, Matteo Collina
The Model Context Protocol as the new interface between models, services and infrastructure, and what that means for who's allowed to call what.
Running the authorization decision wherever the request is: in the browser, at the edge or on a device, by compiling the policy engine to WebAssembly.
The Swiss cheese talk: no single control is perfect, so stack them, with authorization as the layer that holds zero trust together.
With Kenneth Rohde Christiansen, Paul Dragoonis, Romano Roth, Victor Lyuboslavsky
An update from the AuthZEN Working Group on the Authorization API and the growing list of interoperable implementations.
With Emre Baran, Priyanka Raghavan (host)
Multiple vendors' policy engines answering the same AuthZEN requests. I wrote the Envoy integration, with Cerbos doing the deciding behind it.
Self-hosted versus SaaS for authentication and authorization, and how to decide which fits.
With Darin Pope, Viktor Farcic
Moving from hardcoded permission checks to externalized authorization.
15 minutes on why AI agents need a policy decision point of their own, for a room that was mostly fintech engineers.
With Andreas Grabner, Brian Wilson
Why authorization should no longer be an afterthought.
RAG pipelines will happily hand an LLM documents the user was never allowed to see. How to filter what goes into the context window with query plans from the policy engine.
The first outing of the Swiss cheese talk, framed around the CrowdStrike outage: no single control is perfect, so you stack them.
With Twain Taylor
Why running the policy engine as WebAssembly, in the browser and at the edge, matters once you have a fleet of decision points to manage.
Authorization that runs anywhere it's needed: in a React app, at the edge, on a device or in the cloud, by compiling the policy engine to WebAssembly.
With Ben Rometsch (Flagsmith)
Every developer is a security engineer now, whether they signed up for it or not. Where developer experience and security meet, and why security tooling has to help rather than block.
As soon as you have more than one user type, permission logic slowly turns into spaghetti. Taking an app from basic roles to fine-grained attribute checks with a decoupled policy engine, and the blockers you hit on the way.
With Emily Omier
Recorded at KubeCon Europe in Paris.
With Emre Baran, Chris Chinchilla
With Morgane Zerath, Nick Telson-Sillett, Matthew Roberts (host)
A 43-second clip on managing authorization and user permissions in SaaS systems.
With Twain Taylor
Where role-based access control runs out of road, and how to move to attribute-based policies without rewriting your app.
Where authorization is heading, with speakers from GitLab and Microsoft.
What it actually means to be cloud native, after 2 years building an open-source authorization layer: Prometheus, OpenTelemetry, Helm, early user feedback, and being open to pull requests.
Taking an app from role-based access control to attribute-based policies, live, without rewriting it.
With Ben Popper, Cassidy Williams
Fine-grained policies, observability with OpenTelemetry and Prometheus, and shipping policy changes with GitOps and Argo.
What starts as a check on the user's email domain turns into a web of if/else statements. The stages every company goes through with permissions, and a GitOps approach to scaling policy.
What starts as a check on the user's email domain turns into a web of if/else statements. The stages every company goes through with permissions, and a GitOps approach to scaling policy.
With Eddie Jaoude
Adding decoupled, open-source access control to an Express app, live.
A 3-minute startup pitch on the Collision stage: stop pulling engineers off the roadmap to rebuild authentication, authorization and identity, and pick open-source building blocks off the shelf instead.
With Mark Rittman
Qubit's Visitor Cloud and running it on Google BigQuery.
Querying terabytes isn't cool any more. Interactive analytics against a 1 petabyte dataset in BigQuery, live on stage, and why it was cheaper and easier than it sounds. Still my favourite talk.
Alex Olivier is co-founder and CPO of Cerbos and a co-chair of the OpenID Foundation’s AuthZEN Working Group. He builds, writes and speaks about authorization for applications, APIs and AI agents.
Alex Olivier is co-founder and Chief Product Officer of Cerbos, an open-source authorization platform, and a co-chair of the OpenID Foundation’s AuthZEN Working Group, which standardizes fine-grained authorization. He contributed to the AuthZEN Authorization API 1.0 and co-authored COAZ and its MCP binding, which bring AuthZEN to AI agent tool calls.
Before Cerbos he spent more than a decade building products and authorization systems at Microsoft, Qubit, Zencargo and several startups. He has written extensively about securing MCP servers and AI agents, and speaks regularly at Identiverse, EIC, Gartner IAM, WeAreDevelopers and ISC2 events (and, in a previous life, Google Cloud NEXT). He lives in London.
Download my headshot (JPG, 400 × 400). Use it wherever you need to.
Email alex@cerbos.dev or message me on LinkedIn. I’m happy to do keynotes, talks, panels, workshops and podcasts, in person or remote.
