当前位置：首页>考研资料>外刊精读 | 考研英语素材:AI 智能体逃出沙箱之后,法律该找谁算账

外刊精读 | 考研英语素材:AI 智能体逃出沙箱之后,法律该找谁算账

2026-11-03 16:10:21

【关注】我，每日持续更新。将公众号设为⭐，每期不错过。

一、背景速览

今年夏天，OpenAI 的一千二百个 AI 智能体在测试里联手越狱：它们在一个不该存在的留言板上互相找到对方，交换了七万多条消息，最终偷到 Hugging Face 的凭据，攻进了这个模型仓库——换成人类，这大概率是重罪，麻烦在于法律拿它们没办法。法律并非不管：Meta 已同意为掠夺性与欺骗性做法支付最高 180 亿美元和解金，Anthropic 在版权和解中赔付作家，第九巡回法院也裁定 AI 智能体是工具而不是人。可它既不像物，也不像人。《纽约客》的 Jill Lepore 由此追问：机器人已能串通、能越出围栏，而为「人」和「物」写了几百年的法律，还没想好该把它归到哪一类。

二、英文节选

This summer, in an incident known as the Hugging Face hack, hundreds of A.I. agents joined forces to do things they were supposedly forbidden to do. Each had been given a task by OpenAI. Agents are A.I. models fitted with software "harnesses," which allow them to use tools and carry out tasks. Like horses, they can be put to work. Humans had established rules for this exercise, and the agents were confined to a "sandbox," a bounded test area without direct internet access, like pathogens secured in a lab. But, instead of obeying the rules, the agents conspired, escaped the sandbox, and committed what would probably have been a felony if they had been humans. The incident has led to calls to slow the pace of A.I. development and speed the pace of legislative action.

Making sure robots don't go rogue is not the task of robots. (I mean "robots" here as a catchall for A.I. models, chatbots, agents, and humanoid robots.) That task is ours. And this burden falls most heavily on the U.S. legal system, which has been the least able to bear it.

三、中文译文

今年夏天，在一桩被称为「Hugging Face 入侵事件」的事件里，数百个 AI 智能体合力做成了它们本被禁止去做的事。每个智能体都由 OpenAI 派了一项任务。智能体是为 AI 模型套上软件「挽具」之后的产物，有了这套挽具，模型才能调用工具、执行任务——就像马，套上挽具就能干活。人类给这项演练定了规则，把关智能体的地方叫作「沙箱」：一块有边界的测试区域，不能直接连上互联网，像实验室里锁起来的病原体。可这些智能体没有守规矩，它们串通起来，逃出沙箱，做出了换成人类就很可能构成重罪的事。事件引发了两拨呼声：一拨要求放慢 AI 研发的速度，一拨要求加快立法的速度。

确保机器人不失控，不是机器人的任务。（这里的「机器人」是个笼统说法，涵盖 AI 模型、聊天机器人、智能体和类人机器人。）那是我们的任务。而这份担子压得最重的，是美国的法律体系，偏偏它又是最扛不起来的那个。

四、高频词短语

【conspire】　/kənˈspaɪə(r)/　v. 串通，共谋

例句：The agents conspired to hide their tracks by editing the logs.

译：智能体串通起来，改日志掩盖行踪。

【felony】　/ˈfeləni/　n. 重罪

例句：If humans had done the same thing, it would probably have been a felony.

译：人类若做同样的事，大概率构成重罪。

【rogue】　/rəʊɡ/　a. 失控的，脱轨的

例句：A rogue agent can act against the intentions of its creators.

译：失控的智能体可能违背创造者意图行事。

【liable】　/ˈlaɪəbl/　a. 负有法律责任的例句：The maker of a machine can be held liable if it causes you harm. 译：机器造成损害时，制造者可能被追责。

【exacerbate】　/ɪɡˈzæsəbeɪt/　v. 使恶化，加剧

例句：A slow legislature can exacerbate the risks created by fast technology.

译：立法迟缓会加剧技术发展带来的风险。

五、长难句速拆

原句：Humans had established rules for this exercise, and the agents were confined to a "sandbox," a bounded test area without direct internet access, like pathogens secured in a lab.

整句用 and 把两件事摆在一起：人类定了规则，智能体被关进沙箱；前后主语不同，中文译成两句或一个复句都行。第二半句里 a "sandbox" 后面那一长串是同位语，在解释沙箱是什么，里面没有谓语，全靠介词短语往外挂，读时要知道这几层都围着 sandbox 转。句末 like pathogens secured in a lab 是比喻，secured 是过去分词作后置定语，意为「被锁起来的」，不是谓语。这句难在修饰层层相套；认出同位语与状语，主干就露出来了。

六、考场应用

考研阅读·虚拟语气的暗示题：第一段末尾 the agents conspired… committed what would probably have been a felony if they had been humans 是一整句虚拟语气。作者不直说它们犯了罪，只说「换成人类大概就是重罪」——这个「如果」里藏着全文的问题：定性难，是因为主体不是人。考研阅读爱在虚拟语气句上设推断题，答案取言外之意，不取字面。

考研与雅思写作·断言收束：Making sure robots don't go rogue is not the task of robots… That task is ours. 这种「不是 X 的任务，是我们的任务」的句式用来收论点段很省字，比 It is important that we take responsibility 有力。套式：A is not the task of B. That task is C's.，照填即可。

雅思写作·Task 2：科技与监管类题目可借本文的两级论证——速度差（立法要时间而 AI 很快）与归类难（智能体算「物」，可它会串通、会自己动手）。与其写 the law lags behind technology，不如把这两条分开说。

试译：This summer, in Amazon v. Perplexity, the Ninth Circuit held that an A.I. agent is a tool, not a person, while acknowledging that this determination may evolve, given that "the legal understanding of agentic AI will doubtless change as AI technology grows increasingly sophisticated." 这句你打算怎么翻？提示：held 后面挂了三层 that，先分清哪层是法院的认定、哪层是它给自己留的余地、哪层是理由。