A bare "你好" in Chinese triggered Claude's frontier-LLM safeguards and paused the session
menhguin · x · 2026-11-10
A user shared a Claude Code (v2.1.321, Opus 5.5) screenshot where simply typing "你好" in Chinese got flagged under [frontierllm] by the model's safeguards, pausing the session and offering to switch back to Opus 5. The in-app message admits Opus 5.5 has stronger safeguards that can misfire and that improvements are underway — a notable case of over-aggressive safety filtering.
More from Fun
- The mock obituary of 'Carbon': topping leaderboards for weeks before Fable 5.2 replaces it — bindureddy · 2026-11-10
- Developer builds webcam face-controlled browser game in 15 minutes — HashemGhaili · 2026-11-10
- Di Quietly Launches Hospital Companion Service in Beijing, From 126 Yuan per Hour — lipeng0820 · 2026-11-10
- Engineer calls his bank and realizes the voice agent is a model he trained — nathanbenaich · 2026-11-10
- 'You're part of the last generation that will ever have to work on a Monday' — rand_longevity · 2026-11-10
- Independent agent audit finds no evidence Fable stole the 3D aircraft model — victormustar · 2026-11-10