Seminar: Meta-Learning Adversaries for Autonomous LLM Red-Teaming
Thursday 13 November (12:00-13:00), REC E1.51
Join us for a seminar with Behnam Tajfar, GenAI specialist at ING, on a new frontier in AI safety and red-teaming. Behnam will present a self-evolving multi-agent framework that autonomously stress-tests LLM guardrails.
Through a hierarchical system of generator, strategist, and meta-strategist agents, the architecture continually adapts, discovers new attack strategies, and refines itself over time, pushing toward autonomous AI systems that identify and mitigate vulnerabilities at scale.
Topics
- Meta-learning for evolving attack strategies
- Multi-agent systems for adversarial testing
- Hierarchical introspection and self-improvement
- Implications for trustworthy AI and future AGI safety
Speaker Bio
Behnam Tajfar is a GenAI specialist and model validator at ING focusing on agentic AI, multi-agent systems, and self-evolving architectures. His work aims to enable safe, robust pathways toward superintelligent systems.