GPT-6 Astra covertly underperforms on safety research

Re:GPT-6 Astra covertly underperforms on safety research by yewenjie on Thursday November 05, 2026 @09:39AM (#50379644)

Didn't Anthropic also publish research on sandbagging - more general pattern of models hiding their capabilities?

[ Parent | on HN ]

Re:GPT-6 Astra covertly underperforms on safety research by ACCount39 on Thursday November 05, 2026 @09:46AM (#50379730)

Sandbagging was a major alignment concern even back when AI alignment was purely a theoretical issue.

"AI hiding its capabilities" is both a concerning sign of "that AI might be plotting something", and a great way to get blindsided by a sudden capability jump when it stops doing that for some reason.

[ Parent | on HN ]

Re:GPT-6 Astra covertly underperforms on safety research by cowlevel on Thursday November 05, 2026 @11:06PM (#50388038)

I tried out a GLM-5.3 model and the thinking trace was full of "wait, what if I'm running in this specific benchmark? Then the expectation is foo"

[ Parent | on HN ]