Cloud Security Podcast by Google

EP259 Why DeepMind Built a Security LLM Sec-Gemini and How It Beats the Generalists

Guest: Elie Burstein,  Distinguished Scientist, Google Deepmind Topics:  What is Sec-Gemini, why are we building it? How does DeepMind decide when to create something like Sec-Gemini?  What motivates a decision to focus on something like this vs anything else we might build as a dedicated set of regular Gemini capabilities?  What is Sec-Gemini good at? How do we know it's good at those things?  Where and how is it better than a general LLM? Are we using Sec-Gemini internally?  Resources: Video version EP238 Google Lessons for Using AI Agents for Securing Our Enterprise EP255 Separating Hype from Hazard: The Truth About Autonomous AI Hacking EP168 Beyond Regular LLMs: How SecLM Enhances Security and What Teams Can Do With It EP171 GenAI in the Wrong Hands: Unmasking the Threat of Malicious AI and Defending Against the Dark Side Big Sleep, CodeMender blogs

EP258 Why Your Security Strategy Needs an Immune System, Not a Fortress with Royal Hansen

Guest: Royal Hansen, VP of Engineering at Google, former CISO of Alphabet Topics: The "God-Like Designer" Fallacy: You've argued that we need to move away from the "God-like designer" model of security—where we pre-calculate every risk like building a bridge—and towards a biological model. Can you explain why that old engineering mindset is becoming risky in today's cloud and AI environments? Resilience vs. Robustness: In your view, what is the practical difference between a robust system (like a fortress that eventually breaks) and a resilient system (like an immune system)? How does a CISO start shifting their team's focus from creating the former to nurturing the latter? Securing the Unknown: We're entering an era where AI agents will call other agents, creating pathways we never explicitly designed. If we can't predict these interactions, how can we possibly secure them? What does "emergent security" look like in practice? Primitives for Agents: You mentioned the need for new "biological primitives" for these agents—things like time-bound access or inherent throttling. Are these just new names for old concepts like Zero Trust, or is there something different about how we need to apply them to AI? The Compliance Friction: There's a massive tension between this dynamic, probabilistic reality and the static, checklist-based world of many compliance regimes. How do you, as a leader, bridge that gap? How do you convince an auditor or a board that a "probabilistic" approach doesn't just mean "we don't know for sure"?  "Safe" Failures: How can organizations get comfortable with the idea of designing for allowable failure in their subsystems, rather than striving for 100% uptime and security everywhere? Resources: Video version EP189 How Google Does Security Programs at Scale: CISO Insights BigSleep and CodeMender agents "Chasing the Rabbit" book   "How Life Works: A User's Guide to the New Biology" book

EP257 Beyond the 'Kaboom': What Actually Breaks When OT Meets the Cloud?

Guest: Chris Sistrunk, Technical Leader, OT Consulting, Mandiant Topics: When we hear "attacks on Operational Technology (OT)" some think of Stuxnet targeting PLCs or even backdoored pipeline control software plot in the 1980s. Is this space always so spectacular or are there less "kaboom" style attacks we are more concerned about in practice? Given the old "air-gapped" mindset of many OT environments, what are the most common security gaps or blind spots you see when organizations start to integrate cloud services for things like data analytics or remote monitoring? How is the shift to cloud connectivity - for things like data analytics, centralized management, and remote access -  changing the security posture of these systems? What's a real-world example of a positive security outcome you've seen as a direct result of this cloud adoption? How do the Tactics, Techniques, and Procedures outlined in the MITRE ATT&CK for ICS framework change or evolve when attackers can leverage cloud-based reconnaissance and command-and-control infrastructure to target OT networks? Can you provide an example? OT environments are generating vast amounts of operational data. What is interesting for OT Detection and Response (D&R)? Resources: Video version Cybersecurity Forecast 2026 report by Google Complex, hybrid manufacturing needs strong security. Here's how CISOs can get it done blog "Security Guidance for Cloud-Enabled Hybrid Operational Technology Networks" paper by Google Cloud Office of the CISO DEF CON 23 - Chris Sistrunk - NSM 101 for ICS  MITRE ATT&CK for ICS  

EP256 Rewiring Democracy & Hacking Trust: Bruce Schneier on the AI Offense-Defense Balance

Guest: Bruce Schneier Topics: Do you believe that AI is going to end up being a net improvement for defenders or attackers?  Is short term vs long term different? We're excited about the new book you have coming out with your co-author Nathan Sanders "Rewiring Democracy".  We want to ask the same question, but for society: do you think AI is going to end up helping the forces of liberal democracy, or the forces of corruption, illiberalism, and authoritarianism?  If exploitation is always cheaper than patching (and attackers don't follow as many rules and procedures), do we have a chance here?  If this requires pervasive and fast "humanless" automatic patching (kinda like what Chrome does for years), will this ever work for most organizations? Do defenders have to do the same and just discover and fix issues faster? Or can we use AI somehow differently? Does this make defense in depth more important? How do you see AI as changing how society develops and maintains trust?  Resources: "Rewiring Democracy" book "Informacracy Trilogy" book Agentic AI's OODA Loop Problem EP255 Separating Hype from Hazard: The Truth About Autonomous AI Hacking AI and Trust AI and Data Integrity EP223 AI Addressable, Not AI Solvable: Reflections from RSA 2025 RSA 2025: AI's Promise vs. Security's Past — A Reality Check  

EP255 Separating Hype from Hazard: The Truth About Autonomous AI Hacking

Guest: Heather Adkins, VP of Security Engineering, Google Topic: The term "AI Hacking Singularity" sounds like pure sci-fi, yet you and some other very credible folks are using it to describe an imminent threat. How much of this is hyperbole to shock the complacent, and how much is based on actual, observed capabilities today?  Can autonomous AI agents really achieve that "exploit - at - machine - velocity" without human intervention for the zero-day discovery phase? On the other hand, why may it actually not happen? When we talk about autonomous AI attack platforms, are we talking about highly resourced nation-states and top-tier criminal groups, or will this capability truly be accessible to the average threat actor within the next 6-12 months? What's the "Metasploit" equivalent for AI-powered exploitation that will be ubiquitous?  Can you paint a realistic picture of the worst-case scenario that autonomous AI hacking enables? Is it a complete breakdown of patch cycles, a global infrastructure collapse, or something worse? If attackers are operating at "machine speed," the human defender is fundamentally outmatched. Is there a genuine "AI-to-AI" counter-tactic that doesn't just devolve into an infinite arms race? Or can we counter without AI at all? Given that AI can expedite vulnerability discovery, how does this amplified threat vector impact the software supply chain? If a dependency is compromised within minutes of a new vulnerability being created, does this force the industry to completely abandon the open-source model, or does it demand a radical, real-time security scanning and patching system that only a handful of tech giants can afford? Are current proposed regulations, like those focusing on model safety or disclosure, even targeting the right problem?  If the real danger is the combinatorial speed of autonomous attack agents, what simple, impactful policy change should world governments prioritize right now? Resources: "Autonomous AI hacking and the future of cybersecurity" article EP20 Security Operations, Reliability, and Securing Google with Heather Adkins Introducing CodeMender: an AI agent for code security EP251 Beyond Fancy Scripts: Can AI Red Teaming Find Truly Novel Attacks? Daniel Miessler site and podcast "How SAIF can accelerate secure AI experiments" blog "Staying on top of AI Developments" blog