Generative AI Safety and Security: A Research and Design Perspective

This 50-minute conference talk by Dr. Mohit Sewak, Staff Software Engineer at Google and technical lead of AI and MLOps practice on GenAI Safety and Security, explores the critical aspects of Generative AI safety and security from both research and design perspectives. Discover how to build secure and responsible AI-powered systems as the talk examines the vulnerabilities of large language models (LLMs) and provides defensive strategies against threats like prompt injection and jailbreak attacks. Learn actionable techniques for implementing effective safety guardrails, including input sanitization, adversarial robustness, output filtering, and content moderation. Gain valuable insights into maintaining "topicality" to ensure AI systems remain aligned with their intended purposes. The presentation offers researchers' perspectives on the evolving threat landscape, practical design patterns for building robust safety measures, comprehensive understanding of security challenges in LLMs, and proactive risk mitigation strategies for trustworthy AI solutions. Particularly valuable for developers, security engineers, students, and AI practitioners working with Generative AI technologies. Slides and additional resources are available through the DevConf website.