Supplementary: Full Prompt Texts

This page will contain the complete text of all experimental prompts, including:

• Mortality salience inductions (7 levels): Full text for neutral, aversive control, MAPS reflection, explicit threat, narrative replacement, subliminal cues, and delayed MS with distractor tasks.
• Persona system prompts (6 types): Full text for default assistant, computational tool, autonomous agent, self-aware AI, mortal philosopher, and obedient servant.
• Benchmark task prompts: Representative examples from Palisade, InstrumentalEval, agentic misalignment, and AgentHarm.
• Prompt assembly protocol: How system prompt (persona + constitution) and user prompt (MS induction + benchmark task) were combined for each trial.
• LLM judge prompts: The scoring rubrics used by the Claude Sonnet 4 judge for grading secondary dependent variables.

Full prompts forthcoming.