Event ·Official first-hand

OpenAI 发布 MentalHealthBench：评估 AI 心理健康对话的安全性

OpenAI 推出 MentalHealthBench，一个由专家参与构建的基准，用于评估模型在真实心理健康对话中的回复质量。该基准重点衡量 AI 回复是否有帮助和安全，例如面对自伤、焦虑等敏感话题时的表现。它为比较不同模型在心理健康场景下的可靠性提供了统一测试标准。

Current read

OpenAI 做了个心理健康对话评测集，专门测模型遇到自伤、焦虑这类敏感问题时答得稳不稳，各家模型能拿它来横向对比。

11 sources53° AI heatLast updated 2026/9/23 10:00:00

Evidence

11 sources

Primary sourceOpenAI Blog

2026/9/23 10:00:00View source

Related source 1Decoder

2026/9/21 17:44:10View source

Related source 2rohanpaul_ai

2026/9/23 19:28:11View source

Related source 3arXiv: OpenAI

2026/9/24 08:02:48View source

Related source 4VTV Công nghệ

2026/9/21 15:33:00View source

Related source 5宝玉

2026/9/21 20:22:59View source

Related source 6IT之家

2026/9/21 23:27:23View source

Related source 7Sam Altman

2026/9/22 15:06:56View source

Related source 8@OpenAIDevs

2026/9/22 18:14:24View source

Related source 9Gary Marcus

2026/9/22 21:40:24View source

Related source 10theverge

2026/9/23 00:17:17View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story