**OpenAI称Astra为最危险模型,监控难度持续上升** _OpenAI calls Astra its most dangerous model ye
原帖
**OpenAI称Astra为最危险模型,监控难度持续上升**
_OpenAI calls Astra its most dangerous model yet - watching what it does is only getting harder_
> OpenAI正式将即将推出的Astra模型评级为首个具备"关键"网络能力的系统。公司计划通过监控思维链来控制其行为,但Astra的新架构将更多思考过程隐藏于不可读状态,使得安全监控面临更大挑战。
**来源信息**
- **来源**:The Decoder:AI News(RSS)
- **分类**:ai-models
- **发布时间**:2026-09-02 22:20(北京时间)
- **原文**:[打开原文](https://the-decoder.com/openai-calls-astra-its-most-dangerous-model-yet-watching-what-it-does-is-only-getting-harder)