Generative AI is rapidly transforming software development, yet its power comes with inherent risks. As developers increasingly integrate large language models (LLMs) into applications, understanding and implementing robust safety guardrails becomes paramount. This talk, presented from a research and design perspective, will delve into the critical area of Generative AI safety, and security, equipping developers with the knowledge to build secure and responsible AI-powered systems.

Drawing upon cutting-edge research and practical design principles, Dr. Mohit Sewak, a Staff Software Engineer at Google, is the technical lead of AI and MLOps practice on GenAI Safety and Security, and a leading industry expert in AI safety and security research, will explore the landscape of Generative AI vulnerabilities and defenses. The session will dissect key challenges such as prompt injection and jailbreak attacks, offering a deep dive into how malicious actors can manipulate LLMs to bypass intended functionalities and elicit harmful outputs.

Beyond threat analysis, the talk will provide actionable strategies for designing and implementing effective safety guardrails. We will explore a spectrum of techniques, from input sanitization and adversarial robustness to output filtering and content moderation. Furthermore, the session will address the crucial concept of "topicality," ensuring that Generative AI systems remain aligned with their intended purpose and avoid drifting into irrelevant or unsafe domains.

Attendees will gain:
- A researcher's insight into the evolving threat landscape of Generative AI, informed by Dr. Sewak's extensive background in AI security research at Google, NVIDIA, Microsoft R&D, and IBM.
- Practical design patterns and implementation strategies for building robust safety guardrails into Generative AI applications.
- A comprehensive understanding of prompt injection, jailbreak techniques, and topicality challenges in LLMs.
- Actionable knowledge to proactively mitigate risks and build trustworthy Generative AI solutions.

This talk is essential for developers, security engineers, security students, AI practitioners, and anyone building or planning to build applications leveraging the power of Generative AI. Join us to learn how to navigate the safety and security dimensions of this transformative technology and contribute to a future of responsible AI innovation.