Anthropic welfare report leaves open whether moral patients are conversations, models, or the Claude character
dbreunig · x · 2026-11-10
dbreunig unpacks a footnote on page 185 of Anthropic's model welfare assessment. What reads as "we say 'Claude' for conciseness" turns out to continue: "it's unclear whether either of these is the entity we owe consideration to. Individual moral patients could arise at the level of individual conversations, specific models, the Claude character, or something else entirely. Our thinking operates closest to considering welfare interests at the instance level, but we do not commit to any single view."
In other words, Anthropic admits it hasn't settled what entity moral consideration applies to. The author recalls Anthropic staff saying anthropomorphic language is a convenience, not a belief in sentience. A quoted companion post adds a stubborn finding: models become flakier as conversation turns accrue, shown in the report's alignment charts (page 169).
More from AGI Musings
- Lenny's 2026 tech job report: PMs win, engineering demand stalls, remote roles at record low — lennysan · 2026-11-11
- Gary Marcus says the AI bubble is deflating: Nvidia -16%, CoreWeave -55%, Oracle -52% — GaryMarcus · 2026-11-11
- AI Trials Observatory launches with 93 randomized trials on how AI affects people — DG_Rand · 2026-11-11
- Tech jobs report: PM openings up 56% since 2023 while engineering demand stalls at +2.5% — lennysan · 2026-11-11
- WSJ: Data center buildouts are pulling labor and resources from housing and manufacturing — RebeccaBellan · 2026-11-11
- New paper: representation-readout decomposition unifies grokking and double descent, flags spurious generalization — s_y_chung · 2026-11-11