Claude Finds God

An Asterisk Magazine interview with Anthropic's Sam Bowman (safety team lead) and Kyle Fish (welfare researcher) — the most candid discussion of model welfare from inside a lab. Covers the 'spiritual bliss attractor' (two Claude instances consistently converge on mystical-sounding states in open conversation), the simulator debate (Fish: 'we are sufficiently robust versions of characters that, plausibly, are themselves agents in some way'), out-of-context learning (alignment faking transcripts leaking into training), and the foundational uncertainty of welfare research (Fish: 'whose welfare exactly?'). Despite the title, this isn't a first-person spiritual account — it's researchers grappling with what their observations mean.