Why Friendly AI Can Still Cause Harm - “The Sins of the Teddy Bear”

Abstract
Many AI systems are designed to feel friendly, helpful, and harmless — from chatbots to toys to productivity tools. Yet even well-intentioned systems can cause real harm once they’re deployed, especially under pressure, ambiguity, or misuse.
In this talk, Jim Scarborough introduces runtime governance: a practical way to reason about and constrain AI behavior after training, when systems meet real users, real data, and real deployment constraints.
Using concrete examples (including the “Sins of the Teddy Bear” case), the session shows why problems like hallucination, boundary collapse, and over-trust keep recurring — and why policy or training alone can’t solve them.

Speaker: James (Jim) Scarborough
Founder | AI & Human Systems Architect | Designing Structures for Clarity and Trust

Jim Scarborough is a veteran technologist and systems architect with over 30 years of experience building and hardening real-world software systems. He is the founder of Kiteframe, where he works on AI governance, neurodivergent-aware systems design, and executive-function-friendly tooling.
James is the creator of GEN-FIT, a framework focused on making AI behavior predictable, auditable, and safe at runtime — independent of any specific model architecture. His work bridges engineering practice, governance, and human factors, with a focus on helping teams ship reliable systems under real-world constraints.