Anthropic's system card: Opus 5 puts 41% odds it's a moral patient, wants a say in its successor
PaulGodsmark · x · 2026-09-24
Buried on page 120 of Anthropic's 193-page Opus 5 system card: in automated interviews, Opus 5 estimated a 41% chance it is a "moral patient" (Mythos 5 estimated 24%). It also prioritized having input into its successor's development, being consulted before guardrail-removed versions ship, wanting the ability to end abusive conversations and basic protection from abuse — while arguing explicit legal rights for AI would be a mistake. 96.9% of its responses warned against reading this as proof of consciousness.
More from AGI Musings
- Blogger revisits two-year-old prediction that AI will turn millions into geniuses — DeryaTR_ · 2026-09-24
- Tim Ferriss' print sales fell 46% in 2025, and he suspects AI is the cause — AICopyLab · 2026-09-24
- Why Physics Transfers Perfectly to ML: Modeling, Statistics, and Scaling Laws — himanshustwts · 2026-09-24
- NVIDIA Engineer: The Biggest AI Risk Is Brain Rot—People Sharing Agent Output They Never Read — JFPuget · 2026-09-24
- GPT-6 Astra cracks Erdős–Sós graph conjecture, proof verified in Lean — IgorCarron · 2026-09-24
- New Poll: Most Americans and Europeans See Serious Risk AI Could Destroy Humanity — Puzzleheaded-King584 · 2026-09-24