Claude Opus 5.5 reportedly balks at a population ethics problem, researcher suspects 'repugnant conclusion' aversion
anderssandberg · x · 2026-09-28
Researcher Anders Sandberg tried to get Claude Opus 5.5 to analyze a technical problem in population ethics and immediately hit a refusal. He speculates the model balks because it finds the conclusion repugnant (a nod to the Repugnant Conclusion), a notable case study of model behavior on edge-case ethical reasoning.
More from Models
- Users Report Claude Forcing Re-Login Every 2-3 Days, Calling It Exhausting — weswinder · 2026-09-28
- Tokens Are Perfect for Shrinkflation — and Anthropic's Opus Pricing Shift Pressures OpenAI — StewartalsopIII · 2026-09-28
- Benchmark Heaven hits 130k visits in 10 days with cost-capability model rankings — airesearch12 · 2026-09-28
- Which Local Models Sound Least Like Claude? Users Weigh In — Maasu · 2026-09-28
- LLMs Still Stumble on Whether 'December' Contains an X — AlexTensor · 2026-09-28
- Redditor generates a full promo video with one Opus 5.5 prompt in about an hour — spacesheep10 · 2026-09-28