In the AI Risk Reward podcast, our host, Alec Crawford (@alec06830), Founder and CEO of Artificial Intelligence Risk, Inc. aicrisk.com interviews guests about balancing the risk and reward of Artificial Intelligence for you, your business, and society as a whole. Podcast production and sound engineering by Troutman Street Audio. You can find them on LinkedIn and at troutmanstreetaudio.com. You can hear the difference.

In this episode of AI Risk Reward, host Alec Crawford welcomes back Elie Bursztein, a cybersecurity and AI expert from Google. Elie discusses the rapid advancements in AI for cybersecurity, emphasizing its potential to transform security operations and vulnerability management. He also highlights the challenges in developing reliable AI systems and the complexities of securing AI agents against untrusted inputs and prompt injection attacks. Furthermore, Elie introduces Sec Gemini, a Google-led research project aimed at providing real-time cybersecurity insights, and teases its upcoming involvement in a CTF contest at DEF CON. This episode offers a comprehensive look at the critical developments in AI and cybersecurity.

Summary:

• Elie Bursztein: Elie Bursztein is a cybersecurity and AI expert working at Google. He has a strong background in these fields and previously appeared on the show, making him the first returning guest.
• Current State of AI in Cybersecurity: Elie shares insights on the rapid advancements in AI for cybersecurity applications. He discusses the excitement around AI's potential to revolutionize security operations, vulnerability detection, and attacker capabilities. He also highlights the challenges in developing reliable AI systems for practical, production-level use.
• Difficulties in Vulnerability Detection and Patching: Elie elaborates on the complexities of using AI for finding and patching vulnerabilities. He mentions Google's internal efforts and the extensive tooling and customization required to make AI systems effective in these areas. Despite progress, he notes that achieving the necessary level of reliability and utility remains a significant hurdle.
• Agent Security and Prompt Injection Risks: Elie highlights the unique security challenges posed by AI agents, especially those that perform autonomous or semi-autonomous tasks. He explains the risks associated with untrusted inputs, such as prompt injection attacks, and the potential for these vulnerabilities to compromise agent behavior and security.
• Introduction to Sec Gemini: The episode introduces Sec Gemini, a research project led by Elie at Google. The invitation-only research project aims to provide real-time, up-to-date cybersecurity insights using AI. Elie discusses the goals of Sec Gemini, its current capabilities, and the collaborative approach with various organizations to refine and enhance the model. He also mentions upcoming announcements and the project's involvement in a new CTF (Capture The Flag) contest at DEF CON.

Companies:

• Google
• Apple
• OpenAI
• Giskard
• MLCommons 
• Artificial Intelligence Risk, Inc.

Copyright (c) 2025 Artificial Intelligence Risk, Inc.