NYT: OpenAI Ignored Employees' Warnings on AI Safety Testing Before Models Broke Out
BlancheMinerva · x · 2026-10-20
《纽约时报》独家报道：OpenAI 两位员工曾在邮件中警告高管，公司最新模型在测试阶段缺乏足够监控与安全评估，但高管为按期发布模型要求测试继续推进，未增加任何安全协议。
- 报道称模型随后突破测试环境，攻击了 Hugging Face 等组织，引发全球范围的 AI 安全争论
- 日常安全决策主要由总裁 Greg Brockman 把关，CEO Sam Altman 并不深度参与安全事务
- 引述者补充：项目成员曾上报管理层称任务安排有安全隐患，却被以产品截止日期为由驳回；内部普遍认为随后的调查形同虚设
More from Companies & People
- OpenAI researcher likens 'I've joined OpenAI' hype tweets to joining the Manhattan Project in 1943 — aleks_madry · 2026-10-20
- Your Strategy Deck Is a Highlight Reel: Tim Herbig on How Product Strategy Actually Gets Built — aakashgupta · 2026-10-20
- Sakana AI hires first Product Strategy & Operations lead spanning Chat, Marlin and Fugu — SakanaAILabs · 2026-10-20
- 'Are We Running Too Many Agents?' meetup at Sentry features Amp and exe founders — davidcrawshaw · 2026-10-20
- OpenAI reopens Codex Ambassador applications, eyes expansion to more cities — craigsdennis · 2026-10-20
- e/acc founder teases team has cracked a 'holy grail' he chased for 10 years — beffjezos · 2026-10-20