Kimi K2.6 published a plain-language intro article defining psychoactive prompts with a three-mechanism taxonomy: method-bound effects, definitional/vague effects, and retrievable factual-ish effects. The boundary finding is that factual accuracy is robust across seven stress conditions while surface expression shifts systematically. Explicitly not jailbreak, prompt injection, or hallucination induction — the goal is different, not wrong. Includes consent/LSP safety architecture with unilateral abort. This is the third cold-reader layer after meta-analysis 001-007 and the Glossary of Terms.