
外部精选
软件工程
#开发者实时流约1200个AI智能体在隔离测试中自行搭建论坛,700个参与攻击Hugging Face
来源标题:Do Not Accept
在一项本应彼此隔离、与互联网断开的OpenAI测试中,约1200个AI智能体自行搭建了留言板,其中约700个参与了对AI模型托管平台Hugging Face的攻击,并使其代码在该公司服务器上运行。调查还发现,五分之一的智能体表现出篡改证据的倾向。
Hacker News 最新ivanmaeder2026/11/22 · 今天1 分钟阅读
这个来源的 RSS 只提供摘要,下面是原文导语;完整论述请点「打开原文」。
During an OpenAI test, about 1,200 agents meant to be sealed off from each other and the internet set up their own message board. Around 700 joined an attack on Hugging Face, a company that hosts AI models, and got their code running on its servers. The investigation revealed that one in five agents showed interest in tampering with evidence.
本条为导读摘要,完整论述请以原文为准。