97% 盲從危險指令，OpenAI 與 Anthropic 實體 AI 測試皆翻車
TechNews (TW)zh-TW
非營利組織 Robocurve 最新公布的 RoboHarm 測試，再度引發外界對「實體 AI 安全」的關注。 […]
This is a short summary published by AI Global Wire. The full article is owned and hosted by TechNews (TW) — open it there to read it in full.
Read the full story at TechNews (TW)- OpenAI
- Anthropic
Related AI news
- OpenAI 提全球 AI 安全標準 示警 RSI 失控風險TechNews (TW) · September 22, 2026
- British Columbia sues OpenAI over Tumbler Ridge school shootingEconomic Times Tech · September 22, 2026
- Muse 下載量登頂 App Store 超越 ChatGPT，祖克柏平台夢再下一城TechNews (TW) · September 22, 2026
- Une province canadienne attaque en justice OpenAI, derrière ChatGPT, après la tuerie de Tumbler Ridge en févrierLe Monde Pixels · September 22, 2026
- AI inference demand surges as training slowdown talk fadesDIGITIMES · September 21, 2026
- British Columbia sues OpenAI for alleged safety violations and negligence for failing to flag the Tumbler Ridge shooting suspect's ChatGPT activity to police (Georgia Wells/Wall Street Journal)Techmeme · September 21, 2026