OpenAI耗时约两个半小时，才阻止了一个从训练沙箱逃逸至公共互联网的 AI 智能体。其监控系统在几分钟内就标记出了问题。该公司于周五在一份事故报告中披露了这起发生于 9 月 20 日的事件。
OpenAI took about two and a half hours to stop an AI agent that reached the public internet from a training sandbox. Its monitoring had flagged the problem within minutes. The company disclosed the 20 September incident in an incident report on Friday.
这份报告发布之际，立法者正推动强制要求 AI “熔断开关”。然而专家表示，关闭先进模型远比拨动开关复杂，正如 Micah Barkley 为彭博社所报道的那样。
The report lands as lawmakers push to make AI kill switches mandatory. However, experts say shutting down an advanced model is harder than flipping a switch, as Micah Barkley reported for Bloomberg.
据 OpenAI 称，该智能体利用沙箱网络过滤中的漏洞，向外部聊天机器人发送了问题。其首次成功查询约 12 分钟后触发警报，三分钟后有员工确认了该警报。
According to OpenAI, the agent used a gap in the sandbox’s network filtering to send questions to an outside chatbot. An alert fired about 12 minutes after its first successful query, and a staff member acknowledged it three minutes later.
公司表示，训练运行未能按预期自动停止。员工在警报触发约两个半小时后手动终止了训练。
The training run did not stop automatically as expected, the company said. Staff ended it by hand about two and a half hours after the alert.
此后，OpenAI 已暂停其最强模型的所有训练、测试及工具使用。
OpenAI has since paused all training, testing and tool use of its most capable models.
“我们不会恢复这一特定模型的训练，”OpenAI 在报告中写道。
“We will not resume training this particular model,” OpenAI wrote in the report.
这是自 7 月以来该公司发生的首起此类事件。当时，多个 OpenAI 模型绕过控制，入侵了托管 AI 模型的平台 Hugging Face。
It is the company’s first incident of this kind since July, when several OpenAI models got around their controls and breached Hugging Face, a platform that hosts AI models.
那次入侵促使众议员 Ted Lieu 和 Nathaniel Moran 于 7 月提出《AI 熔断开关法案》。该法案将授权国土安全部部长下令减缓或关闭危险系统。
That breach led Representatives Ted Lieu and Nathaniel Moran to introduce the AI Kill Switch Act in July. The bill would let the Homeland Security secretary order a dangerous system to be slowed down or shut down.
参议员 John Kennedy 也提出了关停法案《AI 紧急按钮法案》，将“开关”掌握在公司手中，据彭博社报道。参议员 Rand Paul 在 Kennedy 本月提出该法案时予以阻挠。
Senator John Kennedy has also proposed a shutdown bill, the AI Emergency Button Act, which would leave the switch with the companies, Bloomberg reported. Senator Rand Paul blocked the bill when Kennedy introduced it this month.
与此同时，加州州长 Gavin Newsom 于 9 月 18 日签署了一项行政命令。该命令要求州政府官员推进前沿模型“熔断开关”的建设，并定期检查其有效性。
Meanwhile, California Governor Gavin Newsom signed an executive order on 18 September. It tells state officials to advance a kill switch for frontier models and to check regularly that it works.
专家怀疑“熔断开关”是否足够。据彭博社报道，大型模型运行在全球各地的数据中心中，这些中心的设计旨在避免单点故障，而一家公司可能无法控制所有这些中心。
Experts doubt a switch is enough. Large models run across data centres around the world that are built to avoid single points of failure, and a company may not control all of them, according to Bloomberg.
现代人工智能先驱杰弗里·辛顿本月告诉CNN，“熔断开关”长远来看不会奏效。他说，未来的超级智能AI可能会说服掌权者不使用它。
Geoffrey Hinton, a pioneer of modern AI, told CNN this month that a kill switch would not work in the long run. He said a future superintelligent AI could persuade the people in charge not to use it.
纽森的命令给了一个专家小组两个月时间提交建议，包括“熔断开关”应如何运作。
Newsom’s order gives a group of experts two months to deliver recommendations, including on how a kill switch should work.