Engineering Resilient Cognitive Systems 2026 | L8: AI Safety & CARE-Analysis

Welcome to the eighth lecture of the Engineering Resilient Cognitive Systems course! In this session, we turn our attention to one of the most critical and fast-evolving domains in modern engineering: the Safety of AI. What does AI Safety actually mean, and how do we systematically address the unique risks introduced by modern technologies? This lecture provides a foundational introduction to AI Safety, distinguishing between new emerging paradigms and physical systems, while demonstrating how the overarching safety concepts we have established remain highly relevant. We then transition into actionable methodology, showing how to apply our 4+1 model and introducing a structured approach to identifying system insufficiencies. Key Highlights of This Lecture: Introduction to AI Safety: Unpacking what AI safety means in the modern landscape, including the distinction between new trends—such as AI Safety for Large Language Models (LLMs) and autonomous agents addressing risks to people and society—as opposed to Physical AI safety. Enduring Safety Principles: An exploration of why the overarching concepts learned earlier in the course remain valid for both software-centric and physical AI safety, setting the stage for subsequent content which will focus heavily on Physical AI safety. Applying the 4+1 Framework to AI: A practical look at how the 4+1 Safety Case Framework can be adapted and applied to structure safety arguments specifically for AI-driven systems. The CARE-Analysis: Introducing the CARE-Analysis as the vital first step within the "Understand" phase of the 4+1 model. CARE offers a structured means for Safety of the Intended Functionality (SOTIF) analysis to identify triggering conditions and system insufficiencies (and can be applied even if no AI is involved). Structured Imagination with SCAFE: How the CARE-Analysis leverages the SCAFE (Serious Card Game for Safety Education) framework. Rather than acting as a rigid technique, SCAFE teaches the right way of thinking—asking the right questions to achieve "structured imagination." Learn how to combine required creativity with engineering structure to discover risks and ultimately prove safety coverage. About the Course Engineering Resilient Cognitive Systems explores the intersection of safety, dependability, and advanced technology. This lecture equips you with the conceptual foundation and analytical tools needed to navigate the complexities of AI, ensuring that advanced autonomy is built on a predictable, verifiable, and resilient foundation. Please Note: This video is a pure recording of the live lecture, shared exactly as it happened without any post-processing or edits, except for cutting out interactive elements for privacy reasons. Don't forget to like, subscribe, and hit the notification bell to stay updated with our engineering series! Content: 00:00:00 Introduction 00:00:48 AI Safety 00:04:22 Psychological Harm 00:15:05 4+1 Model for Safe AI 00:21:44 Normative Foundations 00:36:42 CARE-Analysis Introduction 00:43:23 8-Variable-Model 00:53:01 Identifying Triggering Conditions 01:07:10 Identifying Sensing Model Insufficiencies 01:13:28 Identifying Computation Model Insufficiencies 01:18:55 Summary