Pleasure having you here, and as usual, still, I'm the one to do the first talk. The title is a bit boring compared to what I hopefully will present, but it's about identity security for everything, which includes AI, but not only AI.
So, obviously, yes, I will touch AI identity, AI security, but I will also talk about some of the other things, but I want to start with something which my colleague Matthias Reinwald brought up recently. That's from EIC 2019, the summary of the key themes of EIC 2019, seven years ago.
AI, decentralized identity, identity fabrics. And so, I think a little bit of a proof, maybe, that this is the place to learn about the future. I want to start with identity fabric. This is what I talked about last year, so this was 2025, the identity fabric for the 2040s, and what I basically did is I added some aspects.
So, we added AI and smart infrastructures, so we extended the list of target systems. We focused on a more modular approach. We talked about the mesh, so really understanding the identity fabric as something which orchestrates adding orchestration platforms and central authorization platforms to the identity fabric, starting to talk about signal exchange.
We have right now at the SSF, the shared signals framework, as an important thing, and I want to start with the identity fabric and add a little bit of update again here, and to explain that, I'll go back to what triggered the identity fabric when we started it several years ago. We stepped back and asked ourselves, what is the job of identity management?
The job of identity management is delivering seamless yet secure and well-governed access for everyone and everything, and this was in from the very beginning, to everything like services, systems, applications, and not to forget in the age of AI, data. So, this is maybe a bit more clear perspective, very close to our initial identity fabric picture, looking at, again, the signal sharing. I moved it up, because I think it's just where we interact from an identity perspective with everything else, like cyber security services.
We have the structure of capabilities to services to tools, and the capabilities are structured, basically, again, around, I still tend to use the administration analytics or the authentication to 4A. We have the integrations, both bespoke integrations to model legacy stuff to IAM, legacy IAM, and the standards-based integrations, and we have the central layers.
I talked last year about the orchestration layer, the authorization layer, and we should think about the data and relationship layer, a sort of graph we have here, and AI is a bit everywhere, so it's probably not a layer of itself, but probably something which plays more a role everywhere, but it's around identity relations, analytics, AI-powered UX, stuff like that, which we have, and we have the different types of identities, and I thought a lot about the terminology, because I never was super, super happy with non-human, because it's a little bit of a fussy term, and I structured, or we structured it right now around autonomous identities, so the ones that are acting, so to speak, autonomously on their behalf, like an AI agent, which falls into that category, and dependent identities, like workload identities, like identities of things, et cetera, that don't have this level of autonomy, because an AI agent, in a sense, is way more, or an AI identity is way closer to a human identity probably than to a workload identity, and so it's just a proposal, but I think we need to rethink a bit the naming and the structure of it.
So, identity fabrics, here to stay, some updates here, and what I then like to introduce is a cybersecurity fabric, so in a sense, a counterpart to this, and again, looking at a purpose, what is the purpose of cybersecurity? It's secure and resilient data, so information, as an information technology, and operations, the technology in information technology.
We're relatively good on the technology side, we're not so super good on the information side, but it's important in the context of AI, and so when I look at a cybersecurity fabric, the structure, again, is pretty much the same, so we look at the structure of capabilities, because we always must start with the capabilities, not the tool. This is what we need, we need to think about which capabilities do we need, how do we structure them into services, which tools deliver these capabilities.
The capabilities, in this case, you'll see it in a minute, are structured around the cybersecurity framework, basically. We have, again, these different types of actors, in a sense, that do something with systems, which can be the autonomous ones, or the model-dependent, lesser autonomous ones. We have services we protect and data. It's a bit of a misbalance here, so basically data is equally important to all the services, but my PowerPoint skills still are a bit limited, and I didn't want to use AI to create my slides. I did it purely by hand, in this case.
We have integrations, again, what we should think about, like in the identity fabric, how can we deliver, on one hand, an API, a consistent API layer, so a cybersecurity API layer, like the identity API layer, and how can we, on the other hand, share signals? Again, signal sharing is essential, and I really appreciate that we came to standards like the shared signals framework that help us in better sharing signals, telemetry data, et cetera.
As I said, the structure of the capabilities follows the NIST cybersecurity framework with identify, protect, detect, respond, recover, govern.
These capabilities then are mapped to services that could be the SOC service, it could be business application security, it could be third-party risk, it could be data security, or OT security, or AI security, so always a little flexible, so this is not a definite list, always it depends on your environment what you need in services, and this is then delivered by tools which can be more on the platform side of things, or on the modular side of things, depending on whether you lean more towards platformization or best of breed.
It doesn't matter in a fabric because it's about orchestration, it's about bringing these things together, and again, there are some common layers loaded, which is, again, orchestration. Orchestration then also is a functional integration of components, but it's also a signal integration, so there are different ways of integration to look at, and it's also admin flows. I think from the admin flow perspective on the identity side, it's also the user flow, the admin flow. Here on the cybersecurity side, it's probably more the admin flow.
Data integration layer, a seam without functionality is basically sort of an integration layer, and again, AI is everywhere, and when creating fabrics, the logical next step would be AI security fabrics, so when we look at an AI security fabric, the question again is, what is the purpose of that? Why do we need this AI identity, AI security? We need to ensure that AI does what we want, seamless and secure, safe, not going rogue, well-governed, reliable.
This is the job we need to do, and we need, again, something that is about who and what, left-hand, right-hand side, about capabilities and the other thing, so you will see, or you already probably have seen, the structure is quite similar. The who, that could be, again, autonomous and dependent identities, like the humans, the digital twins, the digital co-workers, or autonomous agents and bots, or it could be machine workload entities, whatever else. Capabilities structure into AI security and safety, so safety would be the kill switch or containment.
AI identity, so dealing with the identity, assigning identities, knowing the discovering, governance goes already into discovery, and AI explainability, a very important point. We need to be able to understand and explain what has happened, etc. We have some services, and they are things like discovery services, visibility, observability services, the XLM, so small or large language model security, resource access authorization, authorization, a huge theme, as you all, I think, know, data security, etc.,
and then we have tools, again, to deliver, which could be more the platform or more on the best-of-breed side. I think it's really something which is not put in stone. This discussion is really old, best-of-breed versus suite, we call it, right now we call it platformization, always the same.
Like, we need AI identity management, we need MCP server and other resource server authorization, security platforms for the language models, agent discovery, behavioral analytics, and a lot of other things, and that is to protect the what, it's actions taken by AI and by data produced, delivered, changed by AI. To all the resources, and very important, this entire thing does not stop at an MCP server level. We need to think until the back end, so the MCP system which is behind it, the database which is behind it.
This is where we really need to think through, because if you just end up authorization at a level of an MCP server, etc., it's a very coarse-grained authorization. It's a bit like web access management in the 1990s. We need to go beyond that, the full way down end to end, and this is, to my perspective, built by a mesh of agents, so the technology will be probably, or AI security, be a mesh of interacting agents and a mesh of signals that are used to transport information across all these interacting agents. I think this is one of these fundamental paradigm shifts.
Until now, our thinking is Martin has access to SAP. Now we have situations where Martin does something, and then an agent does something that triggers other agents that do something, and they end up at some resource service. We have no idea where it ends up. It's a mesh. Sometimes a mess. It is something which is very different to everything we did before. It builds on LLMs and SLMs, and all this together, as you've seen, there's a pretty consistent structure.
This is what we need to look at, and these fabrics are related to each other, so there's the cybersecurity fabric with the big picture, the identity fabric that helps us taming the complexity of IAM, and the AI security fabric which helps us securing the new complexity, so we need the one because no security without AI, and no AI without security.
AI identity is a central element of AI security, and identity security brings together these things, but we have different perspectives, different stakeholders, so I believe it makes sense to also have different fabrics that gives us different views, and help us to solve the different problems, but keep in mind all of this is related. One of the big challenges, and this is not a finite list we are facing, is everything we didn't solve in the past years backfires with force now.
So, dynamic authorisation management, we are still not yet there, even while this year is the 50th anniversary of RACF on the mainframe, so we know how to do dynamic authorisation for a while, there's no excuse. We always did technology security firewalls instead of information security. Distributed data-centric security, who has a perfect access control for all this data, warehouse, business analytics, etc., where many systems come together and where you have complex decisions to make, probably no one. Identity relationship management, crafts, etc., still not really good.
Automation, and so we need to work on this to solve the challenges we are facing. The next thing I'd like to look at is where does decentralised identity come in, and I believe this is something we should think about, really, because AI is something that acts autonomously, and in that sense, decentralised. Centralised identity approaches won't fit for this world, so all the things that come from a traditional centralised world, like some of the federation standards, may be part of the solution, but they can't be the full solution.
We need to think about the role of decentralised identities in that game, like, for instance, verifiable credentials, delivering information about guardrails, intent, consent, etc., across the chain, across the entire mesh of agents.
Again, a slide from last year, where I talked about decentralised identities and signals coming together for, I called it back then, policy-based access control. I think it's more than policy-based access control. It's probably a signal-based access control, because we will not be able to describe the policies. If we have 50 or 100 or more signals in a single interaction transaction, no human is able to create that long policy in a written form, so we will need AI to do that, to deal with the signals, which is a bit of a challenge of a recursive distrust, but I think it's very important to do.
And the AI identity placed in this role, where these identities basically can act rather autonomously on these decisions, and authentication, authorisation will change. We will need to be able to identify an agent by the attributes, by the signals, and we can do that. We have everything we need. We need to work on that.
So, traditional IAM is something which tends to fail when we look at the new world of AI identity, AI security. Humans are far too slow for machines.
So, this is fast-moving at scale, doesn't make sense. Static entitlements, dynamic environments, obvious, doesn't make sense. Our traditional direct entitlement assignments, Martin can do that in SAP, doesn't work for the mesh of agents. And all this is non-deterministic. We don't know, we also don't know which agent will knock on the door of a resource server the next minute and ask for what and why. We just don't know it.
So, we work in a different world. We need to rethink.
Also, something from last year, from one of my presentations there, where I said, okay, we need to deliver, organizations need to work on AI. They need to deliver better products, innovation, business process improvements.
So, we have the technology, we use the foundation of data, we don't protect it well, and that is why we need an AI-controlled framework around AI identity, as I said, security, safety, all the stuff which make up the AI security fabric. This is, we need it to enable the business to move forward. And that means we are facing a lot of tectonic shifts. Some of you have been in our workshop in the morning.
So, there are a lot of fundamental changes. I don't want to go into detail. The recording of the workshop will be available, where we discussed all of these tectonic shifts. There will be a lot of research, but it's very clear that the entire thing is fundamentally changing, as I've said, non-directed, non-deterministic, a lot of new challenges, multi-tier authorization, etc. For the fabric, the next thing we will deliver, and this is just a bit of an idea where this is heading, we will, in the next step, deliver also our reference architecture for that.
This is something I did a bit ago, looking at the pillars of AI governance, AI security, AI identity, and there are quite a lot of elements we have that we need to bring together to build this AI security fabric in our organizations. Some of that is here, some not. Will be an interesting evolution, and it will lead to new market segments. There will be new technology areas. My colleague Jonathan defined some of them. I will talk later this week about AWAP and ATDR, so agent visibility and observability, not just visibility but also solving things, platforms.
I will talk about agent threat detection response. Jonathan will talk about other areas. There will be new tools, there will be a lot of things, and there's a lot of stuff to do. That's what I want to end up with.
We need, as an industry, as a community, work together, and to quote a German comedian, I can't take care of everything, make the most of it. Thank you. Congratulations on being bang on time. What else? A tough act to follow, but we do have one question here. You packed a lot in there, tectonic shifts and the like, but the question here is, what is the most practical first step to move from this fragmented visibility that everyone's got to a coherence governance model across all the identities that you mentioned? I think that's important.
When you look at all the new types of technologies, when you look at the tectonic shifts, we don't have solutions for all of that. We can take technical measures for a lot of them. We talked about this in the workshop, but to start with, I think there are two logical points. The one is discovery, so understanding which agents do we have. We can't protect, we can't govern what we don't know, so discovery is a logical starting point.
The other, so to speak, edge is the access to the resources. In front of the resources, to do as much and as good an authorization as we can do at that time. It will evolve with more signals, with different types of authorization mechanisms, but it's still the logical starting point because this is a sort of a control point. We can get a grip on all the agents.
Great, thanks very much. Moving swiftly ahead to our panel, please take a seat. You're on the panel.
