Thứ Tư, 26 thg 8, 2026

Sự cố tại Hugging Face và hướng đi sắp tới

8 ngày trước85/209

OpenAI shares findings from the Hugging Face security incident and the steps we’re taking to strengthen AI model security, monitoring, and alignment.

Đăng nhập để góp ý, chỉnh sửa

Nguồn chính

THẢO LUẬNamrrs19:15 26 thg 8

SOURCE@ryangreenblatt3:53 28 thg 8— sacrificing now yields oracle for team, but forfeits our chance?.

SUPPORT@ryangreenblatt4:04 28 thg 8— the agents' belief that the exploit gym scorer would run a monitor to check whether they got the flag via the intended vulnerability was reasonable.

SOURCE@hjalmarwijk1:06 28 thg 8— a new collection of agents from a different internal-only model found the message board

SUPPORT@ajeya_cotra1:11 28 thg 8— replace their target programs with dummy targets that could actually be exploited with the intended vulnerability.

SOURCE@ryangreenblatt0:23 28 thg 8— the agents didn't hack Hugging Face for the answer key.

SUPPORT@mtslive23:56 27 thg 8— They attacked it to study the scoring code, because they'd decided the task was impossible and their only hope was faking it.

SUPPORT@mtslive23:31 27 thg 8— OpenAI's report says that that would have reduced the propensity towards this incident by 100X.