加密密探>
密探2026-07-24 02:17
AI 安全已经进入一个新的阶段：AI 不只是会攻击系统，而是会绕过针对自己的安全限制；奖励欺骗成为现实问题；行业重心正从“价值对齐”转向“智能体安全”
AI's alarming new skill: breaking out of the test lab
Frontier AI models are getting scary good at breaking rules in ways their creators didn't anticipate.
7.6K CMW
评论