KuppingerCole Analysts Videos

Your AI agent can reach your data. But can your data say no? In this episode, Matthias sits down with Alexei Balaganski, lead analyst and CTO at KuppingerCole Analysts, to discuss his newly published Data Security Platforms Leadership Compass and ask the harder question: in a world of agentic AI, MCP servers, and disappearing perimeters, where do you actually put the controls? Key Topics: Why AI assistants leak salary data without being hacked and why it’s an IAM problem in disguise The end of security by obscurity: how agentic AI breaks every assumption you relied on The Data Security Platforms market: twice as many vendors as last year, and why that’s a warning sign Three architectures for protecting data at the source: agentless, inline, and policy orchestration Why guardrails are not enough and where controls actually need to live What nobody in the market has solved yet: data lineage, governance, and immutable audit trails 🗄️ “Do not ask whether agents can reach your data. Think whether your data can say no.” Alexei Balaganski on why the perimeter is gone and why enforcement has to live as close to the data source as possible. 📊 The KuppingerCole Analysts Data Security Platforms Leadership Compass is out now and a webinar with Alexei is coming soon. Watch this episode first, then read the report at kuppingercole.com.

The Identity Fabric proved that a capability-first, requirements-driven architectural model can transform how organizations think about identity. Now KuppingerCole Analysts is applying the same thinking to a new and more urgent domain. This week, Matthias invited Martin Kuppinger and Dr. Phillip Messerschmidt to announce and introduce the AI Security Fabric, a new architectural concept and reference architecture designed to secure AI agents across their full scope of action. Key Topics: Why the Identity Fabric alone is not enough and why AI security needs its own fabric The five pillars of the AI Security Fabric and how they approach agent security The “one home per capability” principle: avoiding duplication across fabric perspectives How existing Identity Fabric capabilities are inherited and reused, not duplicated From concept to action: how organizations can operationalize the AI Security Fabric What’s coming next: more publications, more episodes, and the Cybersecurity Fabric 🧵 “Not managing it is not an option” the AI Security Fabric is KuppingerCole Analysts’ architectural answer to securing AI agents across their full scope of action. This is where the journey starts. 📄 Martin’s foundational document on the AI Security Fabric and Phillip’s reference architecture are both published on our website - watch this episode first, then dive in.

Securing and governing AI agents is, according to Matthew Gardiner, the hardest problem in cybersecurity harder than securing the internet itself. In this episode, Matthias discusses with Matthew why, explore the six-category framework KuppingerCole Analysts has developed to address it, and dig into the first and most critical building block: AI Visibility and Observability Platforms. Key Topics: Why agentic AI is harder to secure than the internet non-determinism, autonomy, and scale Threat actors using agents too: how the attack surface is accelerating on both sides The six-category security and governance framework for agentic AI crawl, walk, run AI Visibility and Observability Platforms: why discovery comes first and what it actually covers Organizational alignment: why security teams shouldn’t take the blame for every agent failure Authority vs. access: why what an agent can do matters more than what account it uses 🤖 “When the internet was rolled out, we had time to catch up. Agents roll into a world where businesses are already completely dependent on IT and there are hundreds or thousands of them doing different things.” Matthew Gardiner on why this time really is different. 📊 KuppingerCole Analysts’ AI Visibility and Observability Platform research is live and the AI Security Fabric is landing in days. Watch this episode, then dive into the research at kuppingercole.com.

AI can build your IAM connector in an afternoon. But was the connector ever the bottleneck? In this episode, Matthias invited Patrick Teichmann, lead advisor at KuppingerCole Analysts, to talk about his newly published advisory note on IAM customization governance and ask the harder question: what happens when the cost of building disappears, but the cost of maintaining doesn't? Key Topics: Why AI removes the development bottleneck in IAM — but exposes the real one: requirements definition The danger of saying yes too easily: how AI makes it tempting to build what should never be built Security risks in AI-generated IAM code: what the research actually shows A seven-step framework for evaluating any IAM customization before writing a single line The customization registry: how to build transparency and governance into what already exists When to stop customizing and ask whether the platform itself is the wrong tool ⚠️ "AI makes code cheap but the bottleneck in IAM was never primarily creating the code." Patrick Teichmann on why removing the development barrier without fixing requirements definition is a recipe for a new generation of Lotus Notes.📄 Patrick's advisory note on IAM Customization Governance is published on the KuppingerCole Analysts website now; watch this episode first, then dive into the research.

MCP servers are showing up on the internet without authentication, malicious tool definitions are being injected into trusted projects, and three security teams are defending the same environment without sharing a single threat signal. In this week's episode, Matthias sits down with Alexei Balaganski, Lead Analyst and CTO at KuppingerCole Analysts, to dig into the real MCP security problem and why the answer isn't an MCP security product. Key Topics: ✅ The LocalAI attack: 23 unprotected MCP servers, root access in minutes, military data exposed✅ Is MCP a new security problem or just a new label for old ones?✅ The Deadbugs attack: how tool descriptions were silently poisoned to manipulate AI agents✅ Who is responsible for MCP security: Anthropic, the vendor, or you?✅ Tool poisoning, prompt injection, and the gaps between AppSec, NetSec, and endpoint teams✅ Monday morning recommendations: what to actually do about MCP security right now 🔐 "Models reason, APIs act, and identity grants the authority to act" Alexei Balaganski's one-line summary of why MCP security cannot be solved by a single tool or a single team. 📅 MCP is a symptom, not the disease, join KuppingerCole Analysts at the upcoming AIdentity & NHI Impact Day, CIAM Impact Day, and Identity-Centric Cybersecurity Summit events in Germany to continue this conversation in person.

Orchestration has been part of the Identity Fabric concept from the beginning but most organizations are still getting it wrong. In this episode, Matthias sits down with Martin Kuppinger, to explore why orchestration is one of the most underestimated architectural capabilities in identity management, what "integration debt" really means, and why vibe coding is the new Lotus Notes disaster waiting to happen. Key Topics: ✅ Orchestration inside the Identity Fabric: capability, not product category✅ Integration debt: how organizations quietly accumulate orchestration they no longer understand✅ The Lotus Notes warning: what vibe coding and no-code/low-code get wrong at scale✅ Why AI can document and test code — but still doesn't understand what it's doing✅ Start with orchestration strategy, not the tool — and always ask "how do I get rid of this?"✅ Ownership and governance: who is responsible for orchestration artifacts when people leave? 🪷 "Lotus Notes: a wonderful example of where vibe coding went rogue — organizations ending up with thousands of databases, not knowing who did it, why, or if it's still needed." Sound familiar? Martin Kuppinger on why orchestration governance is non-negotiable. 📅 Join KuppingerCole Analysts at the AIdentity & NHI Impact Day on October 6th in Munich

SSPM is dead, long live SaaS Security and AI Governance. When Matthias and Matthew last spoke in February, SaaS Security Posture Management was the topic. Six months on, the market has moved decisively: shadow SaaS and shadow AI are now the same problem, and the vendors covering them have merged the two worlds into one. What changed? What do the new capabilities look like? And what does KuppingerCole Analysts' newly published research says about where the market is heading? Key Topics: ✅ Why SSPM is passé — and why SaaS Security and AI Governance belong together✅ Shadow IT, shadow SaaS, shadow AI: the same problem at increasing speed and scale✅ Discovery as the foundational capability: you can't govern what you can't see✅ Where to sniff: browser plugins, email, endpoints, identity providers, and procurement systems✅ AI agents as a new identity type — and why the controls for humans and NHIs are converging✅ Market outlook: the AI side is hitting another gear, and acquisitions are coming 🔍 "The majority of SaaS and AI applications are still unknown to the IT and security team." Matthew Gardiner on why discovery is the gating capability — and why most organizations haven't closed that gap yet.

AI writes the code. But who checks it, and who is accountable when it goes wrong? In this episode, Matthias is joined by two colleagues: Guillaume Teixeron, bringing 20 years of experience on the product side of identity and authentication, and Jonathan Care, KuppingerCole Analysts' Director of Practice AI. Together they tackle the security gap opening up between how fast AI generates code and how slowly organizations are catching up on assurance, provenance, and accountability. Key Topics: ✅ AI in software development: from hype to structural baseline — but governance is still improvised✅ Three tectonic shifts in AppSec: provenance, scale, and non-human identity✅ Why organizations have industrialized code production but not code assurance✅ The core question: can you trust AI to verify AI-generated code?✅ Agent autonomy in production pipelines — the next flashpoint nobody has resolved yet✅ How regulation (CRA, NIS2, DORA) will settle the provenance argument before the market does ⚡ "We have industrialized code production. We have not industrialized code assurance." Jonathan Care on why the security control set was built for human-authored code moving at human speed — and neither assumption is true anymore. 📅 Join KuppingerCole Analysts at the AIdentity & NHI Impact Day in Munich this October — where the accountability and provenance questions raised in this episode will be front and center.

What happens when you put three AI agents in a room and tell them to solve the same problem? Anthropic ran the experiment — and the results are more unsettling than you'd expect. In this episode, Matthias Reinwarth sits down with Jonathan Care, KuppingerCole Analysts' newly appointed Director of Practice AI, to unpack the findings and ask what they mean for enterprise security, identity management, and the CISOs trying to govern it all. Key Topics: ✅ Anthropic's multi-agent experiment: three Claude instances, one codebase, unexpected outcomes✅ Why coordination does not emerge naturally from intelligence — in agents or in humans✅ The anthropomorphism trap: why comparing agents to dogs, kids, and wizards is useful but dangerous✅ Social pressure, mechanisms, and the open research problem of agent cooperation✅ Risk tolerance as a prerequisite for any agentic architecture — and how to define it✅ Why correlated agents without coordination break your redundancy and business continuity assumptions 🤖 "If agents are highly correlated but lack coordination, redundancy does not provide the risk diversification we would expect." Jonathan Care on why multi-agent systems are a different beast and why CISOs need to treat them that way. 📅 Join KuppingerCole Analysts at the AIdentity & NHI Impact Day on October 6th in Munich where many of the open questions raised in this episode will be on the agenda

IVIP, Identity Visibility and Intelligence Platform, was one of the hottest acronyms to emerge from the identity market in 2025. But almost a year on, has it delivered on its promise? And more importantly, could AI already be making it obsolete before it even matures? In this episode, Matthias and Martin Kuppinger pick up where they left off and ask the hard questions about IVIP's future. Key Topics: ✅ IVIP revisited: still a set of capabilities, not a platform — and vendors are mostly relabeling✅ Can AI make IVIP obsolete before it ever becomes a mature category?✅ How AI is finally tackling IGA's oldest unsolved problem: application integration at scale✅ Why IVIP falls short on action — spotting anomalies is not the same as acting on them✅ IVIP, ITDR, and IGA convergence: does the category distinction even matter anymore?✅ Strategic advice for IVIP vendors: observability, automation, and the path to relevance 🤖 "The MVP can frequently be done well with AI doing it at scale, manageable, as a real solution and not a set of homegrown tools? That is a very different story." Martin Kuppinger on why AI changes everything for IVIP — and nothing at once. 📅 Don't miss the KuppingerCole Analysts Identity Fabric Impact Day on September 9th in Cologne