Key Takeaways
- Chris Betz has moved from VP of Infrastructure Security to CISO of Google Cloud, expanding his customer‑engagement focus, especially around AI‑driven threats.
- He advocates a three‑pillar framework for AI‑powered security: expert → harness → model, stressing that expertise and a well‑designed harness outweigh reliance on a single top‑performing model.
- Google Cloud employs a multi‑model strategy, leveraging diverse AI models (including Gemini, Anthropic’s Project Glasswing, and others) to improve coverage, reduce token cost, and adapt quickly to evolving threats.
- Offensive security remains central: red‑team projects like Naptime, Big Sleep, and CodeMender automate vulnerability discovery and fixing, giving defenders contextual advantages attackers lack.
- To harden the software supply chain, Betz backs initiatives such as Alpha‑Omega, promotes memory‑safe languages (e.g., Rust), and uses AI to replace low‑value open‑source dependencies, reducing patch churn.
- Human‑in‑the‑loop processes—automated AI‑assisted code review coupled with expert security oversight—ensure AI‑generated code meets quality bars and that security fixes reach production.
- The biggest current challenges are the rapid pace of AI‑driven innovation and securing the right talent to scale defenses; Betz views this as both a technical and a people problem.
- Looking ahead, he envisions a future where vulnerability counts shrink dramatically, and urges fellow CISOs to seize the heightened board‑level awareness of security to drive lasting change.
Background and New Role
Chris Betz officially assumed the title of Chief Information Security Officer (CISO) for Google Cloud after serving as the company’s VP of Infrastructure Security. While he describes the shift as primarily a title change, it broadens his mandate to include deeper customer engagement, particularly on AI‑related threats and the broader security transformations underway in the industry. Betz notes that he has been leading the Google Cloud CISO team for roughly a year, spending significant time with engineering groups across Google Cloud and within the CISO organization itself. His responsibilities now encompass the security of Google Cloud’s underlying infrastructure and guiding customers to achieve optimal security when they build on top of Google’s platform. Reflecting on his career, Betz traces a thread of global‑scale service from his U.S. Air Force beginnings through intelligence work at the NSA, senior security roles at Microsoft, Apple, CenturyLink, AWS, and Capital One, positioning his move to Google Cloud as a natural continuation of his mission to protect large, interconnected ecosystems.
Vision for AI in Cyber Defense
In his first interview since becoming CISO, Betz emphasized that traditional security approaches are insufficient for an AI‑first threat landscape. He argues that organizations must develop, improve, and share capabilities that allow AI to defend against AI. Betz highlights his advantageous position at Google Cloud, which hosts a frontier AI lab, enabling him to shape the conversation around AI‑driven defense. He observes that virtually every CISO he speaks with now integrates AI into their daily workflows, causing their roles to evolve rapidly. Consequently, one of his immediate challenges is to provide security teams with greater AI capacity so they can move faster—a problem he describes as enjoyable because it reflects progress rather than stagnation.
Multi‑Model Approach and Harness
Betz outlines a three‑pillar framework for building AI‑powered security defenses: the security expert, the harness, and the model—in that order. He stresses that using a single AI model is inadequate; instead, Google Cloud employs multiple models, including its own internal models and participation in external efforts such as Anthropic’s Project Glasswing. Different models excel at different tasks, and even the same model can yield varied answers on identical problems, making a multi‑model strategy essential for robustness and coverage. This approach also helps optimize cost: by diversifying models, Betz’s teams can achieve the best performance per token while keeping expenses low. The second pillar, the harness, refers to a tool or layer that wraps an AI model, supplying context, data, threat modeling, and explicit guidance on where and how the model should focus—such as when hunting for vulnerabilities. Building an effective harness requires skilled engineers who can encode complex logic and domain knowledge. Betz advises that if only two pillars can be chosen, expertise and a strong harness are preferable to a top‑tier model lacking either, because human judgment and contextual framing amplify the utility of even modest models.
Offensive Security and Red Team Initiatives
Continuing the philosophy that “the best security defense is good offense,” Betz describes Google Cloud’s investment in red‑team activities that discover and remediate vulnerabilities before adversaries can exploit them. He mentions conversations with CISOs across France, Germany, and the UK, all of whom report seeing more than ten times the volume of vulnerabilities compared with previous years. To stay ahead, Google maintains multiple red‑team units: some focus exclusively on Google Cloud, contributing to projects like Naptime (automated vulnerability discovery), Big Sleep (scaling that discovery), and CodeMender (automated fixing). CodeMender, launched the previous fall, follows a “bring‑your‑own‑model” philosophy, allowing teams to employ Gemini or any other suitable model. Betz underscores that defenders possess unique advantages—rich contextual data, full stack insight, and ownership of the environment—that attackers lack, making these AI‑assisted offensive tools especially powerful. He anticipates an exciting period of advancement over the next six to eighteen months as these capabilities mature.
Open Source Supply Chain Improvements
Addressing the growing concern over vulnerabilities in open‑source software, Betz highlights Google Cloud’s commitment to strengthening the software supply chain. He points to the Alpha‑Omega initiative, a Linux Foundation‑sponsored program that pools funding to drive security improvements in critical open‑source projects. Internally, Betz is evaluating where open‑source components are truly needed and where they can be supplanted by AI‑generated code, particularly aiming to transition from memory‑unsafe languages (C, C++) to memory‑safe alternatives like Rust. His team has identified many open‑source dependencies where only a tiny fraction of the code is actually used; in such cases, retaining the full library adds unnecessary risk and patch‑management overhead. By leveraging AI‑based software development to create purpose‑built, minimal replacements, Google Cloud aims to reduce dependency churn and overall attack surface while maintaining functionality.
Human‑in‑the‑Loop and Engineering Partnership
Betz is adamant that AI‑generated code must not bypass rigorous security scrutiny. To mitigate the risk of introducing new vulnerabilities via AI, Google Cloud employs automated AI‑assisted code review processes alongside traditional security code reviews on every piece of code that ships. This human‑in‑the‑loop strategy ensures that any AI‑produced suggestions are vetted by expert security personnel before release. Moreover, he emphasizes the indispensability of a tight partnership between security and engineering teams. A security team that identifies flaws but lacks the engineering collaboration to implement fixes in production will fail to protect customers effectively. By fostering close communication and joint ownership of remediation efforts, Betz believes Google Cloud can translate vulnerability discoveries into reliable, timely patches that maintain both security and development velocity.
Challenges: Speed of AI and Talent
When asked about the most pressing challenge, Betz identifies the breakneck pace at which AI drives innovation and the difficulty of applying AI defenses to AI‑generated threats as the foremost technical obstacle, a problem likely to persist for the next few years. He couples this with a people‑centric challenge: finding and retaining the right talent capable of innovating at massive scale. No CISO can tackle these issues alone; success depends on building teams that are comfortable wielding novel AI tools, developing new tools, and adapting their workflows week over week. Betz views the talent gap as equally critical as the technical one, noting that attracting individuals who can thrive in this fast‑moving environment is essential for sustaining effective defense.
Outlook and Advice to Fellow CISOs
Looking forward, Betz expresses optimism that the vulnerability landscape could shrink dramatically within the next two years. He reflects on decades of battling software flaws and suggests that the combination of AI‑powered discovery, automated fixing, supply‑chain hardening, and stronger engineering‑security collaboration may drive the number of exploitable vulnerabilities to a fraction of today’s levels. To fellow CISOs, his advice is simple yet urgent: seize the current moment. Public awareness of vulnerability risks has never been higher, and business leaders and boards now grasp the significance of security challenges in ways previously unseen. Betz urges security leaders to use this heightened awareness to engage executives, secure necessary resources, and guide their organizations toward lasting, resilient security postures. By acting now, CISOs can translate heightened concern into concrete, long‑term improvements that protect both their enterprises and the broader digital ecosystem.