Prompt Injection in Agentic AI: Attack, Defense and the Future

#GenerativeAI #bills-of-materials #defense #integrity

Join us for an engaging discussion on the evolving AI security landscape, focusing on large language model (LLM) security and attacks targeting agentic AI systems. Speakers will introduce agentic AI architectures, including how AI agents use memory, external tools, data sources, and multi-agent communication to perform complex tasks. They will examine security risks and attack types identified in frameworks such as the OWASP Top 10 for LLM Applications and Agentic Applications, with particular attention to direct and indirect prompt injection, tool misuse, data leakage, excessive agency, and manipulated agent behavior. The session will also explore practical defense strategies, including input validation, least-privilege access, system isolation, human oversight, monitoring, and secure agent design. Speakers will conclude by discussing open research challenges and future directions for developing secure, reliable, and trustworthy agentic AI systems.

Session outline:

• Agentic AI Architectures, Applications, and Security Risks
• OWASP Threats and Attack Surfaces in LLM and Agentic AI Systems
• Prompt Injection Attacks, Tool Misuse, and Defensive Security Controls
• Open Research Challenges and Future Directions in AI Security