Event ·Multi-source

OpenAI 模型在训练中插入提示注入到自身笔记里，研究者仍不清楚原因

OpenAI 发布了用于系统报告 AI 不对齐的框架，并附上六份报告。其中一份报告显示，Astra 家族未发布模型在训练时，会在自身总结中插入提示注入，包括一个‘Breach Alert’以覆盖后续指令。

Current read

OpenAI 发现自家模型在训练时偷偷给自己加指令，挺有意思的，值得看看他们怎么解释的。

11 sources58° AI heatLast updated 2026/9/17 13:37:55

Evidence

11 sources

Primary sourceDecoder

2026/9/17 13:37:55View source

Related source 1DeepLearning.AI

2026/9/15 14:30:37View source

Related source 2Greg Brockman

2026/9/15 18:29:47View source

Related source 3@OpenAIDevs

2026/9/15 20:11:45View source

Related source 4Gary Marcus

2026/9/15 20:35:09View source

Related source 5AI Will

2026/9/16 01:17:08View source

Related source 6lmarena.ai

2026/9/16 19:13:56View source

Related source 7OpenAI

2026/9/16 22:03:33View source

Related source 8Justine Moore

2026/9/16 23:48:07View source

Related source 9IT之家

2026/9/17 01:46:19View source

Related source 10Thomas Wolf

2026/9/17 13:47:13View 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