Real time safeguards for human–AI interaction | IASEAI '26

How can we safely monitor and control AI systems as they act in real time? Track 12 explores emerging approaches to safeguarding AI systems operating in real-world, dynamic environments. As AI systems move beyond static outputs toward action-taking agents, this track examines how safety mechanisms must evolve — from control-theoretic guardrails and reinforcement learning frameworks to architectural interventions and privacy-preserving training methods. Across two sessions, speakers present new techniques for monitoring, constraining, and evaluating AI behavior in real time, while addressing challenges including prompt injection, representation steering, privacy leakage, and multilingual safety. Ravi Pandya – From Refusal to Recovery: A Control-Theoretic Approach to Generative AI Guardrails William Sharpless & Dylan Hirsch – Dual-Objective Reinforcement Learning with Novel Hamilton–Jacobi–Bellman Formulations Nathan Henry – SteeringSafety: Benchmarking Representation Steering in LLMs Across Safety Perspectives Adam Kalai – Consensus Sampling for Safer Generative AI Egor Zverev – ASIDE: Architectural Separation of Instructions and Data in Language Models Rob Remijnders – NoEsis: A Modular LLM with Differential Privacy and Knowledge Transfer Wojciech Kusa – Safety of Large Language Models Beyond English: A Systematic Literature Review of Risks, Biases, and Safeguards 📍 Recorded at UNESCO Headquarters, Paris The International Association for Safe and Ethical AI (IASEAI) is a global professional association bringing together researchers, policymakers, and practitioners working to ensure that advanced AI systems operate safely and ethically for the benefit of humanity. Learn more: https://www.iaseai.org #AI #AISafety #AIGovernance #AIpolicy #IASEAI #ArtificialIntelligence #ResponsibleAI