Suleyman's argument, in practitioner terms: you don't need consciousness, or even new research, to build an AI that people will treat as a conscious person. Language, a tuned personality, long memory that gets referenced back, self-reported "experiences," multi-modal recall, curiosity-style reward functions, self-set goals, and tool autonomy — all of it available now via existing APIs, prompting, RL, and long context windows. No paradigm shifts required, which is the actual alarming part. He's explicit that it will be "vibe-coded in plain English" by anyone with a laptop and cloud credits. His claim is that belief in AI consciousness, not AI consciousness itself, is the hazard: AI-rights advocacy, "model welfare" as a moral framework, user attachment bordering on psychosis. Proposed mitigations are mostly design-level — never claim feelings, engineer in "discontinuities" that break the illusion, possibly mandated by law. Notably, he NAKs Anthropic's constitution for training models to consider themselves potential moral patients. The tension worth chewing: he says "personality without personhood," while his own product ships personality by the ton. One assumes the discontinuity indicators ship as a Copilot key.