Securing AI: From Models to Agents

In Episode 8, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about what big tech companies are actually doing for AI security. They discuss real-world practices, challenges, and lessons from industry AI deployments.

In Episode 12, host Honghui Xu talks with Shiva Shrestha about building a secure Physical AI systems under real-world constraints. They discuss how to balance security, usability, cost, and performance in practical AI deployments.

In Episode 11, host Honghui Xu talks with Tahmid Imtiaz Imu about backdoor attacks in LLM agents. They discuss how hidden triggers can affect agent behavior and why detecting these attacks is important for building safer AI systems.

In Episode 10, host Honghui Xu talks with Emama Nahid about model-level defense for LLMs. They discuss how quantum-inspired ideas can help improve model robustness and security.

In Episode 9, host Honghui Xu talks with Md Abdullahil Oaphy about detection-based defense for multimodal LLMs. They discuss how to detect risks across text and image inputs, and how these defenses can make multimodal AI systems safer and more trustworthy.

In Episode 7, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about how AI security defenses are evolving. They discuss why prompt-based tricks are not enough and why stronger architectural security is needed for real-world AI systems.

In Episode 6, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about multi-agent and system-level attacks. They discuss how attackers can exploit interactions between agents, tools, and systems, and why stronger system-level defenses are needed.

In Episode 5, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about indirect injection and data poisoning. They discuss how these hidden attacks can compromise LLM systems and why they are difficult to detect and defend against.

In Episode 4, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about attacks on LLMs. They discuss prompt injection, jailbreaks, data leakage, and agent hijacking, and explain why these attacks matter for real-world AI systems.

In Episode 3, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about why LLMs are fundamentally vulnerable. They discuss how these risks come from the way LLMs are trained, designed, and used in real-world AI systems.

In Episode 2, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about the Five-Layer Stack as a new way to understand AI security. They discuss how different layers of AI systems create different risks and require different defenses.

In Episode 1, host Honghui Xu talks with Shiva Shrestha and Kazi Sharif about how AI security changes as LLMs evolve into agents. They discuss why security is no longer just about text, but also about tools, actions, data, and real-world systems.