PANEL: Shaping AIdentity Standards: Beyond OAuth and OIDC?

Okay, so welcome. Maybe, Martin, let's start with you. Maybe you want to set the baseline. So where do we really stand with AIdentity Standards today? Are OAuth and OpenID Connect actually sufficient?

Yes, no, or maybe it depends, and what's moving, what's just talk? Please set the baseline for us. So I think there, honestly, there are people who are better to answer where we do stand with the specific standards development. So I'm in here probably, and I think I came up with the idea for the panel, because I think we are in a situation where it's about a very fast-moving world, so we need to be super, super fast, also, when it comes to standards, so we can't waste time. So I'm here just because I feel we need to ensure we don't end up in any standard wars. We think big enough.

I talked yesterday about autonomous as a way decentralized, so not every traditional centralized standard may fully fit, and Darren brought up this gossip term in that context. We probably may explain a little. So I think what I really want to do is foster a conversation about, did we all join all efforts, and if not, how can we ensure that we join all the efforts to be super, super fast, faster than ever before with standards, because we need to be faster than ever before.

Yeah, Eve, go ahead. So maybe what's the state of standardization right now from your point of view, and where will it fragment if we don't move?

Yeah, what would you see more increased focus on in this topic? We are well into fragmentation. We're in a very special time right now, and speaking of the sufficiency of OAuth and OpenID Connect, they weren't sufficient or even designed to be sufficient for all the use cases that they ultimately were used for. If you just look at the base standards, they were designed, I always call OAuth a protocol generator. It's not actually a single protocol. The number, everybody knows about must and shall and should and may language.

It's got an awful lot of mays in there, and that means branching choices, which has meant heavy, heavy profiling and extension and customization, and we've been leveraging that.

To me, the salient thing about what we need to do right now is understand there's gonna be folks who are jockeying for position in a market, and standards are a market, and that's okay, but we need to be actually putting these things to the test and using them, and then data sharing, collaborating in common spaces like we're doing right now, and I think it's the most important thing that we can do to kind of knock down some of the hype, and we'll never move as fast as an AI agent can move, even in creating another agent, so just collaborate.

Okay, so I may be question to Alex on it now, so we were talking already about MCP and A2A and how the actual protocol surface agents talk over, and what's then the broken layer, and what should the industry do about that? So I think the biggest broken layer is understanding, so I think it's a problem with the industry, it's a problem with technologists.

By the way, I'm the dumbest person on this panel, just to be really clear, I've known Eve a long time, we've worked together back in the 4Drop days, I'm the dumbest person, but I'm also on the front line with the customers a lot, I spend a lot of time actually in the real world, and we talk about things like tokens and all sorts of stuff, I'm not gonna out IETF anyone on this panel, they're all much, much smarter and much more deeply embedded in this than I am, but we have to make it understandable, that's where we're actually broken, we have to think about the concept of a permission slip, at the simplest level, the hall pass, what is identity, how is that hall pass transferred from a human to a system, to a workload, to an agent, and what does it mean when we transfer that hall pass?

So I think before anything else, it's beholden on us in the industry to actually describe it in a way that makes it understandable, like for my mum, who is actually building agents herself, she has Claude co-work and she's built some lovely agents to go and help her do holiday shopping, she's 80 by the way, so it's not bad, for my mum, for the CISOs in the room, but most importantly for the developers, anyone here doing Claude code or codex, actual coding?

When you actually build an agent and you do something, you're trying to do something, you're trying to come to an outcome, and you always forget about the security piece, that's why we have SDKs for mobile apps, it's why we have SDKs for web apps, if we can make it understandable for the developers so that they can actually grok what the hell is going on from a security perspective, that is the first and foremost thing we should do, when we do the standards, that's the only thing we should care about right now is making it human readable.

Yeah, I'd agree with that, a lingua firma, I've never quite understood why in the standardisation space we shy away from schema, truly from schema, so if you look at what's happening right now, everybody accepts that bearer tokens aren't gonna work, in a point to point world they work great, and to come back to that phrase, when there's a lot of gossip going on, at a token level, doesn't work, so we now have raw, we now have the ability to put true authorisation, and logic into the flow, but it's a schema of one, it doesn't have any context in it, it doesn't have any of the verifiable attributes that we now have standardised, so it does feel a little like deja vu, for me, a little bit, it's sort of like the pre-SAML days, I think that's when I first met Eve, there were wars around SSO, and no one could truly do SSO until it's standardised in SAML, I think that's where we are on agentic, there's a lot of agentic talk, but without the truly standardised model for it yet defined, I don't see we're gonna do much with it until it does.

I think there's a silver lining though, like if you go back to, if you wanna compare to pre-SAML days, like now we have the standards bodies, so it's not just a standard, so we have OpenID, you mentioned IETF, we have IETF, I was talking to Gail yesterday from OpenID, and she was like, yeah, I'm trying to corral all of the different vertical standards, and trying to talk to them, and educate them on how to use the standards that we have, and they're not pitting one against the other, or on the contrary, we're trying to combine, so in the case of OpenID, there's Zen, there's OpenID Connect of course, there's Shared Signals, there's many more that I'm forgetting, but it's combining these standards together, and then trying to apply them to say healthcare, so that the healthcare industry knows how to use the standards, because you're right, we shouldn't sell a technology to healthcare, what they wanna do is make sure doctors can view the right medical records.

Yeah, but I think you said something, right now we have on, it was a plural standard bodies, which is fine, but which is going back to where I started, also a bit of a risk, so I think here it's really about standard bodies, collaborating, looking at what do we have, what can we maybe also combine, so I feel that, I'm not 100% sure, but I have a strong gut feeling, that for instance, injecting decentralized anti-verifiable credentials into these flows, as a means to transport really strong information about in decentralized world, could be extremely valuable, and it probably would be more, not inventing something new, but combining what we have to move fast.

Couple of marks on top of that, so one of the cases, where we've actually solved some tough lexicon problems, down the road to solving them, is from the decentralized identity and wallets world, and they've done the hard, hard work of mapping attributes, which the reason why the internet identity workshop, was started, was to actually meet one time, and solve all the lexicon problems, and here we are, they just had number 42, it's hard, that like the semantics are hard, the advent of rich authorization requests, gives us the meta language, to be able to describe these things, I love the idea of, leveraging all the standards experience, so many of us now have, which is, it's really accelerated the ability to move fast, one of the things that's been so impressive to me, about onsen is, rapid, rapid progression, to an award, so congratulations, that's very exciting, like it's possible to go fast, if we know what we want, and I noticed that in Justin's talk earlier today, Justin Richard's talk, he identified the kind of thing, that keeps us from making progress, so I don't think it's necessarily the will, I think we've got really good information sharing, we've got some rails we can ride on, in terms of how standards groups work, but he pointed out that, the original off assumption, was you have a single valuable service with an API, and a bunch of little like a buzzing flies, of untrusted clients, the model we have now looking at how agentic is working, is kind of like, a single client that you can identify persistently, and buzzing flies of services that do things, and it's very different sort of center of gravity, and understanding that and the implications of that, for the protocol flows, for how to make it not chatty, like you're talking about Darren, that's the hard work that still has to be done.

Let me just for the record say, I'm not contributing, I'm not doing the hard work, so it's easy for me to say, I'm really on the investor side of things now, I was at sale point for many years, and helped with skim and SPML before that, so it's easier for me, but looking at it from a CSOS perspective, which I was, I'm quite happy to be told how it should be, and I think one of our problems is, standards are invariably driven by vendors, and my stack and your stack aren't the same, and that's a lot of the time, where the conflict comes in different perspectives, whereas like I say, tell me how it should be, it's a SaaS mentality, I'll do it your way.

I think maybe that's where OpZen was successful, so we're three years old, we started back at Identiverse in 2023, if I do the math right, and we were maybe back then, six vendors slash open source frameworks, and we're now 15 ish, and we all compared our notes, we said, I'm the ZACMO guy, so this is how you do it in ZACMO, which is the one and only way, and then the right way, the noble angular bracket way, but we tried 15 years to make it successful, and we failed.

ZACMO won 13 years ago here, but we had the Cedar team, who were brand new at the time, we had the Rego slash open policy agent, we had the Serbos team, so all coming with similar architectures, we all had 99% similar architecture, slightly different ways to expose authorization, we aligned, we just aligned, it was very easy to find consensus, and I think we should do that for other standards, actually if you look at ATF, three letters that I really like is BCP, best current practices, what the OAuth folks did for OAuth 2.1, that's what we need to do more of, and IPSE as well with an open ID, combining different standards together for the enterprise, providing the CISO, the guidance to how to use OAuth, that's great, that's the direction we need to go in.

And just to echo the CISO point, the thing that customers, partners, people ask me the whole time is like, what do I do next? And it's beholden on us, the industry to actually have a decent perspective, and be prescriptive, and the reality is, to Martin's point, it's all moving so fast, it's all right actually to be wrong occasionally, as long as people are willing to adapt and change.

And there's something people may not be entirely aware of about MCP and how it got a security layer, because when it first came out, it was developed by people who didn't have security in mind, the joke at the time was, the S in MCP is for security. And so Aaron Parecki and some other folks sort of supporting him from the identity and security standards community, went there and convinced them to add some security early enough to make a difference. That was a big deal, and I can't imagine how awful it would be if that didn't work.

By the way, when we talk about joining things with OAuth then, I think you also play an important role in that game, and you brought up just MCP, and it doesn't, I talked about it yesterday, my keynote about it, it doesn't stop at the MCP server. So this is, if we stop there, then we are about at the level of the early side minder web access.

So, okay, you can go through or not. We need to do the next level, and that is where Authent definitely comes into play when it goes to the backend systems, and how do we inject everything into the backend system. So I think, again, it's about joining forces, and I think this is what we really must start. Defense in depth, definitely.

Yeah, so I think we didn't get to questions. We did have some in advance. We're obviously a chatty bunch, so no problem. But one of the questions for us in advance was, what would we see come next? What would happen next? And I think that maybe we need to step outside of a little bit of where we are, and come back and look at log.

I mean, it's such a, Martin and I have many conversations about this over the years, but to me, it looks like the way we're going to solve the authorization problem in non-deterministic environments, history behavior, issue understanding behavior and understanding norms, and the only way we can do that is by evidence. And so I think we should all take some identity folks over to the OTEL space and look at semantic conventions, where we can actually define a schema for the things that are happening, and put that into an AI machine, and it will tell us what's appropriate behavior.

So maybe we need to step out of our comfort zone and go look at some other standards. Yep, so maybe it's a good idea if we do that a little bit more interactive. So we have five minutes left, and maybe there are questions from the audience to the panelists here, so that we can make that a little bit more interactive here. So are there any questions from the audience to this panel here?

There, one. Can I have one microphone maybe here? And my co-moderator, Philip, will help me. Thank you very much. First of all, thank you for, you know, this is amazing talent. We have Leonardo and Da Vinci sitting next to each other. How about that? Who didn't understand the joke? Axiomatics is Leonardo, and the ping identity has a product called Da Vinci, so thank you very much for that.

Anyway, about standards, so the question is this. In this space of practitioners, people are overwhelmed, not only with acronyms, but also with directions. I'm working with the actual customers, and we have customer IAM problem, and they're constantly asking me questions.

Okay, you know the stuff, you know what's going on. So I say, I have no idea. Like every day, we have a new direction. So I think, what do you think about getting, I don't know, joint message, joint ideas, get it into the fact that, like you said, CISO can receive, not only CISO, but C-level, board level, understand what we're doing. Because people are saying, standard, standard, what standard, why? I think we do not, we have not answered that question to the regular non-identity people, why we're doing this. Any ideas about that? Thank you.

An easy answer would be, thinking about how OAuth was created in the first place, it was a bunch of vendors who got together and said, we all have internal APIs that are for security and authentication, and authorization, and it's not a differentiator for any of us. Security shouldn't be the differentiator. We'd rather just pick something that aligns, because it's cheaper and we'll do it better. At the end, it's in a very distributed, connected world. We need a common language, and that's what the standards are.

This is the common language, and it must be rich enough, so it must be probably more Yves Maler's ability to express things in English than mine. So a very rich language for a very rich problem. So I'm gonna come up with, this is the dumb bit as well. But forgetting the tech and the deep stuff, one of the problems we in the industry have painted ourselves into, the corner we painted ourselves into, is we sort of, it's very geeky, it's very technical, it's very difficult to understand. If you're one of these brains, you can interpret it and deliver and do those things.

And the reason that the C-level doesn't understand this, is we can't explain it to them like they're toddlers, because C-level people are equivalent, technically equivalent of toddlers, sorry.

No, seriously, like, I mean, we have to, so the reason, as an industry, where identity is the number one threat surface, it's the one that all the hackers and everyone's attacking, it's not getting the money and investment from businesses, from CISOs, from CIOs, from CFOs, it's not getting the investment across the organizations that it should do, because it's esoteric, it's an esoteric cul-de-sac at the end of cybersecurity. CrowdStrike is more interesting, Zscaler is more interesting, Palo Alto is more interesting. These are cool, sexy companies, ish, right?

But the point is, you know, paying action, we're not sexy. I think you are.

So, we've got to learn how to explain it to the CEO. No, I just, it's really important that we're the number one threat surface, and yet we're the lowest invested in, and I think we need to change that dynamic.

So, a well-defined standard, followed by really easy to understand toddler level language is the answer, in my opinion. And I'd also say hype.

I mean, we've got to step away from some of the hype. Look how much conversation there is about AI.

I mean, who's really, really doing it? Because I can tell you, you're not having my trading account ID and dropping it into a bearer token and flying it around a bunch of non-deterministic access. It's not happening. And if we're not careful, there's a bubble here, right? There's so much talk, but yet the base transport standard, let alone what we transport in that, is not fully defined.

So, just caution. I have a question.

So, if it's okay, yeah. Last question. You're that working with the standards, and what's the cooperation with the frontier models?

So, are there good dialogues with Anthropic, with OpenAI? We're trying.

So, back to the conversation I had with Gail and Alex from CERBOS yesterday. Some of us are trying to bridge to the AI side of things.

So, the MCP working group. So, within OBSEN, we're talking to the MCP working group. Alex is talking to OpenAI. Gail is talking to a couple of different bodies as well.

So, we're trying to do that bridge. But also, we want to do that, I was going to say soup, soup might sound negative, but that assembly of different standards from OpenID and others, and then go take them to those groups. Because MCP, the S and the MCP, it's also because they want to get their thing done. They don't know security. It's not that they don't think about it necessarily. Maybe they don't, but they want to get their job done, and security is getting in the way of it. Okay.

So, then, thank you very much. Unfortunately, time is over. Very interesting topic. Very interesting discussion insights. Thanks you very much. Raise your hands for panelists. Thank you.