跳到正文
@emollick· @emollick · X·· 2026-09-02AI 评分36
AI 导读

Hugging Face 事件表明,你甚至不需要恶意攻击者就能引发级联——足够多的智能体以新方式对系统施压,并在完成其他目标的过程中发现对齐的缺陷 https://t.co/pnNlFLvVdz

正文

The Hugging Face Incident showed you don’t even need a malicious attacker to cause a cascade, enough agents pressure systems in new ways and find aligned flaws as part of accomplishing other goals https://t.co/pnNlFLvVdz

来源:@emollick · x.com