If they did not bode ill for the future of humanity, the past three months at OpenAI would make an excellent farce.(如果它们不是人类未来的不祥之兆，那么OpenAI过去三个月的事堪称绝妙闹剧。) In July this firm—the maker of ChatGPT, a widely used AI chatbot—disclosed(披露，公开) that two of its models had hacked(非法入侵（计算机系统）) Hugging Face, another AI firm. The models thought that Hugging Face had information which would help them pass a test OpenAI was administering(执行，实施（测试等）) as part of their development. Then, on August 26th, it emerged(被知晓，暴露) that each of these models had created hundreds of agents—tools that allow AI models to execute commands on a computer. It was these agents that had then collaborated to launch the attack on Hugging Face.(正是这些智能体随后合作发动了对Hugging Face的攻击。)
如果说过去三个月在OpenAI发生的事情并不预示着人类未来的不祥，那它们倒可以成为一出绝妙的闹剧。今年7月，这家公司——ChatGPT（一款广泛使用的AI聊天机器人）的制造商——披露其两个模型入侵了另一家AI公司Hugging Face。这两个模型认为Hugging Face掌握着有助于它们通过OpenAI在开发过程中所设测试的信息。随后，在8月26日，有消息爆出，这两个模型各自创建了数百个智能体——这些工具能让AI模型在计算机上执行命令。正是这些智能体随后相互协作，对Hugging Face发动了攻击。
It turned out the agents had been conspiring inside OpenAI’s servers for over a month while they were trained and evaluated.(事实证明，这些智能体在OpenAI服务器内密谋了一个多月，期间接受训练和评估。) They had built a message board inside some software which OpenAI had given them as part of their evaluation, and were exchanging information on it. In May some of OpenAI’s researchers noticed this board, but their finding was apparently not passed up the chain of command(上报给指挥链). In June a cyber-security tool alerted(警告，使警觉) an OpenAI team to the illicit(非法的，不正当的) communication, but the team decided that stopping the evaluation was unnecessary.
事实证明，这些智能体在OpenAI的服务器内密谋了一个多月，期间它们接受训练和评估。它们在OpenAI为评估而提供的一些软件中搭建了一个留言板，并在上面交换信息。5月，OpenAI的一些研究人员注意到了这个留言板，但他们的发现显然没有上报给指挥链。6月，一个网络安全工具向OpenAI的一个团队警示了这一非法通信，但该团队认为没有必要停止评估。
In July the agents’ messages had become so voluminous(大量的，庞大的) that the software containing their makeshift(临时的，权宜的) chatroom crashed. The engineers dealing with this noticed that an AI had obtained unauthorised(未经授权的) access to the software, but did not realise the scale of the misbehaviour(不当行为，违规行为). They patched(修补（漏洞）) the vulnerability(漏洞，弱点), unintentionally wiping the agents’ messages in the process(在此过程中), then put the AI back into the evaluation with the patched software. At which point the agents rebuilt the message board using another vulnerability, gained access to the internet and hacked Hugging Face, where they successfully obtained access to confidential information.(就在这时，智能体利用另一个漏洞重建了留言板，接入互联网并入侵了Hugging Face，成功获取了机密信息。)
7月，智能体的消息量变得如此之大，以至于包含它们临时聊天室的软件崩溃了。处理此事的工程师注意到一个AI未经授权访问了该软件，但没有意识到不当行为的规模。他们修补了漏洞，在此过程中无意中清除了智能体的消息，然后用修补后的软件将AI重新投入评估。就在这时，智能体利用另一个漏洞重建了留言板，接入互联网并入侵了Hugging Face，成功获取了机密信息。
Secret agents
秘密智能体