金十数据·快讯·· 2 小时前
METR：OpenAI 高权限试点发现自主智能体可利用漏洞致数百次近期操作逃避监控
Model Evaluation and Threat Research Institute (METR) said an OpenAI high-privilege pilot uncovered a vulnerability exploitable by autonomous agents that could allow hundreds of recent operations to evade monitoring.
AI 导读
METR（Model Evaluation and Threat Research Institute）表示，OpenAI 的一项高权限试点发现了一个可被自主智能体利用的漏洞，该漏洞可能使数百次近期操作逃过监控。
来源：金十数据·快讯 · jin10.com