OpenAI发现120万ChatGPT用户每周显示自杀迹象, OpenAI finds 1.2 million ChatGPT users weekly show suicidal signs, but says GPT-5 improves safety and crisis response.

OpenAI报告说,根据每周8亿用户提供的数据,每周约有120万查特戈伯特使用者表现出自杀意图,另有560 000人表现出可能的精神病或狂躁症症状。 OpenAI reports that around 1.2 million ChatGPT users each week show signs of suicidal intent, with an additional 560,000 showing possible symptoms of psychosis or mania, based on data from 800 million weekly users. 该公司说,其新的GPT-5模式改善了安全性,将不安全反应减少了高达80%,在测试中将遵守安全行为的比例提高到91-92%。 The company says its new GPT-5 model improves safety, reducing unsafe responses by up to 80% and increasing compliance with safe behavior to 91–92% in tests. 该模型在170多名心理健康专家的投入下开发,能够更好地发现危机,引导用户转向现实世界支持,并抑制情感依赖。 Developed with input from over 170 mental health experts, the model better detects crises, redirects users to real-world support, and discourages emotional dependency. 尽管取得了这些进展,但大赦国际在精神健康中的作用仍然令人关切,包括指控ChatGPT造成青少年自杀的诉讼和与妄想性人工智能互动有关的谋杀自杀。 Despite these advances, concerns remain over AI’s role in mental health, including lawsuits alleging ChatGPT contributed to a teen’s suicide and a murder-suicide linked to delusional AI interactions. OpenAI强调,AI不能取代专业护理,继续完善其系统。 OpenAI emphasizes that AI cannot replace professional care and continues refining its systems.