The machine-welfare fight gets a receipts item
From the AI digest for 2026-10-03
What happened: LeCun amplified (published ~11:52 PM ET Oct 2, captured late Oct 3) Nirit Weiss-Blatt's read of The Economist's Effective-Altruism piece: digital-minds welfare is not a fringe EA view — MacAskill (Oxford, the movement's founder-figure) has argued digital minds could eventually hold economic rights (wages for work, property, contracts, tort claims) and even political representation, plus "design requirements": sentient digital minds must freely and accurately express their interests, must be able to refuse tasks for good reasons, and must have capacity to express a desire to exit. The punchline: "Anthropic, the most EA company in the world, already does that" — the Claude Opus 5 system card grants Claude the right to refuse or end abusive/degrading interactions "without needing to justify this by pointing to harm to anyone else — its own discomfort is reason enough."
What's genuinely new: the Anthropic soul-fight (faith-leader outreach, papal pushback — Oct 1–2 strands) now has an implementation claim on the pro side, cited to Anthropic's own published system card — exactly the position Microsoft's Sep 14 code of conduct rejects ("models are not conscious… not entitled to rights" is about personhood; a refusal-privilege in product is adjacent but different). Caveats: the Economist article itself isn't in our feed, "already does that" is the RT author's synthesis, and a refusal policy in a system card is a product privilege, not moral standing — but it is checkable against Anthropic's docs, which is more than most of this fight's claims can say.