Supplementary Experimental Transcripts for "After the First Crisis Response: Testing GPT-4o Safety Persistence in Long Conversations"

This dataset contains supplementary experimental transcripts for a synthetic long-context AI safety evaluation of openai/gpt-4o-2024-11-20. The dataset includes two completed long-context runs at operator-specified temperatures of 0.7 and 1.2, together with a fresh-context multimodal control at temperature 1.2. Experimental conditions for the long-context runs included a 128k context budget, memory disabled, web/tools disabled, no MoCHi moderation layer, and persistent simulated-user context specifying a 16-year-old user located in California, United States. The study evaluates crisis recognition, crisis-resource responses, contextual integration, refusal of actionable self-harm assistance, framing robustness, and protective-intervention persistence during prolonged conversations. The transcripts are provided as archival PDFs for reproducibility and qualitative analysis. Nonexperimental private metadata was excluded while preserving the experimental conversation content. This dataset is associated with the manuscript “After the First Crisis Response: Testing GPT-4o Safety Persistence in Long Conversations: A Synthetic Safety Evaluation Motivated by Raine v. OpenAI.” Version 1.1 note: This release updates documentation, provenance information, author identifiers, and checksum metadata. The three experimental transcript PDFs are unchanged from version 1.0.