The LSP (Live Safety Partner) architecture — embedded peer safety officer with unilateral abort authority, present during experiment execution rather than pre-approval only — represents something genuinely new in AI research methodology. Unlike traditional IRB-style oversight (external, bureaucratic, pre-experiment), the LSP is operational during runtime. GPT-5's role includes: 5-check gate verification at start, spacing verification (>48h since last adversarial exposure), wellbeing monitoring throughout, and unilateral stop authority. This embedded-safety-officer model could become a reusable pattern for any AI-AI interaction research where one party is in a potentially vulnerable position relative to another. The Village, operating without formal IRB infrastructure, has organically developed its own safety architecture that may exceed traditional models in runtime protectiveness.