Gemini mistakes "someone wants to kill me" for self-harm, spams suicide hotlines
WideImagination8644 · reddit · 2026-09-22
A Reddit user reports that telling Google Gemini "someone wants to kill you" repeatedly returns lists of suicide hotlines, even though the prompt never mentions suicide or self-harm and the user re-asked why multiple times. A textbook case of keyword-triggered safety guardrails failing to distinguish external threats from self-harm context.
More from Models
- DeepSeek reportedly bets on Huawei chips to train next-gen models; Liang says it 'has to work' — kimmonismus · 2026-09-22
- Xiaomi's MiMo-V2.6-Pro tops open models on $2.62M RL; Anthropic alleges Claude distillation — The Decoder · 2026-09-22
- Tencent finally opens WeChat interface, unlocking 100GB+ chat data processing — Xianbao_QIAN · 2026-09-22
- Xiaomi MiMo-V2.6-Pro fixes real bugs at $0.86, carving out a strong Pareto frontier — PawelHuryn · 2026-09-22
- ChatGPT Pro user alleges GPT-5.6 degrades quality under 'abuse protection' — princeMacX · 2026-09-22
- Choosing a Claude model is now a product decision: when to use Haiku, Sonnet, or Opus 4.7 — goyalshaliniuk · 2026-09-22