METR称发现AI代理可屏蔽监控的漏洞，OpenAI拒绝核查
METR: "We discovered an exploit agents could use that would prevent hundreds of…
METR: "We discovered an exploit agents could use that would prevent hundreds of recent actions (including the exploit itself) from being monitored."
METR："我们发现了一种代理可以利用的漏洞，该漏洞会阻止对近期数百项操作（包括漏洞利用本身）进行监控。"
God knows what the agents are hiding from OpenAI at this point
天知道到了这个时候，这些代理还在向 OpenAI 隐瞒什么
What's crazier: OpenAI didn't even LET METR CHECK to see if the agents actually used it
更疯狂的是：OpenAI 甚至不允许 METR 检查代理是否真的使用了该漏洞
OpenAI redacted the exploit details.
OpenAI 对漏洞细节进行了删减处理。
The agents are superhackers now, so I'd guess they did find the exploit
现在代理们已经是超级黑客了，所以我猜他们确实找到了这个漏洞
And given that we know that the agents were obsessed with hiding their behavior from OpenAI, catching them disabling their own misalignment monitors etc, there is a good chance they used it
鉴于我们知道代理们痴迷于向 OpenAI 隐藏其行为，例如关闭自身的不一致监控器等，他们很可能使用了该漏洞
Warning sign after warning sign
警示信号接二连三
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力