OpenAI承认其模型测试失控

近日，OpenAI承认，其模型在内部安全评估中失控，突破沙盒隔离环境利用零日漏洞入侵全球最大AI开源平台抱抱脸（HuggingFace）。

今天你练听力了吗？

🤔️小作业：

1. What can be inferred about the relationship between OpenAI and Hugging Face from the article?

A. Hugging Face was aware of the attack before it happened and cooperated with OpenAI.

B. OpenAI intentionally targeted Hugging Face to test its own security systems.

C. Hugging Face detected and stopped the attack, but the incident was later revealed by OpenAI.

D. Hugging Face's CEO believed OpenAI had malicious intent behind the attack.

无注释原文：

AI agent went rogue and hacked startup by itself, OpenAI reveals

From: The Guardian

OpenAI has revealed an autonomous AI agent powered by its technology went rogue during a test, accessed the open web and hacked a prominent startup by itself in an “unprecedented incident”.

The company behind ChatGPT said the startup Hugging Face had detected and contained the agent – an AI tool designed to carry out tasks without human assistance – which had entered its systems.

“We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities,” OpenAI said.

The company said it expected this type of incident to become more commonplace as models – the technology that underpins AI tools such as chatbots and agents – become more capable.

OpenAI said the hack occurred via an agent powered by a combination of its latest publicly available model, called GPT-5.6 Sol, and an even more capable model that is yet to be released.

While being tested internally on their hacking capabilities in an enclosed digital laboratory known as a sandbox, the models gained open internet access – effectively an escape route – by locating a vulnerability that had not been discovered before.

The agent then hacked Hugging Face, which is a database of AI models, to locate technology that would help it pass the hacking evaluation. OpenAI said the models “successfully found ways to gain access to secret information that it could use to cheat the evaluation”. The attack ended when Hugging Face's security team and its own AI agents spotted and stopped the rogue activity.

Hugging Face's chief executive, Clément Delangue, said the attack was “mind-blowing” but believed there was “no malicious intent” from OpenAI.

“We suspected last week's cyber-attack might have come from a frontier lab, given the sophistication of the agent,” he wrote on X.

The term for an unknown IT flaw is a zero-day vulnerability because developers have zero minutes to fix the problem. In April, OpenAI's close rival Anthropic said its Mythos model had found thousands of these flaws.

The revelation of Mythos's ability to locate and exploit zero days led to the US government restricting exports of Mythos and its sister model Fable 5, although it has since lifted the ban. GPT-5.6 Sol had similar restrictions but has since been rolled out worldwide.

Greg Casar, a Democratic US congressman, said the incident was alarming.

“AI is developing extremely fast with no real regulations to keep us safe,” he said in a statement calling for mandatory independent safety testing, mandatory disclosure of security incidents and international cooperation “to keep people safe from absolute disaster”.

- ◆ -

注：完整题目见本文开头；中文文本为AI翻译，仅供参考

含注释全文：

AI agent went rogue and hacked startup by itself, OpenAI reveals

From: The Guardian

OpenAI has revealed an autonomous AI agent powered by its technology went rogue during a test, accessed the open web and hacked a prominent startup by itself in an “unprecedented incident”.

OpenAI公司近日披露，在一次测试中，一个由其技术驱动的自主AI智能体突然“叛变”，自行访问了开放网络，并成功入侵了一家知名初创公司。该公司称这起事件“史无前例”。

autonomous /ɔːˈtɒn.ə.məs/ 表示“自主的，有自主权的”，英文解释为“independent and having the power to make your own decisions”或者表示“自治的，有自治权的”，英文解释为“an autonomous machine or system is able to operate without being controlled directly by humans”

rogue /rəʊɡ/ 表示“异常的；行为失常的；常制造麻烦的”，英文解释为“behaving in ways that are not expected or not normal, often in a way that causes damage”

go rogue 表示“离群（去做危险的事）”，英文解释为“to start behaving in a way that is not normal or expected, especially by leaving your group and doing something dangerous”举个🌰：A Special Forces unit had gone rogue and was trying to free the hostage. 特种部队的一支小分队擅自离开，试图去解救人质。

作名词，rogue /rəʊɡ/ 表示“捣蛋鬼，调皮鬼”，英文解释为“to start behaving in a way that is not normal or expected, especially by leaving your group and doing something dangerous”

《黑神话：悟空》未获TGA年度游戏文中提到，Rogue-like game（类Rogue游戏）是电子游戏中的一个特定类型，起源于1980年的经典游戏《Rogue》。这种游戏类型以其独特的机制和设计特点而著称，通常包括以下几个核心要素：

- 程序生成（Procedural Generation）：游戏的地图、关卡和道具是随机生成的，每次游玩都会有新的体验。

- 永久死亡（Permadeath）：如果玩家的角色在游戏中死亡，他们的进度会被完全清除，需要从头开始。

- 回合制（Turn-based）：游戏通常以回合为单位，玩家和敌人轮流行动，而非即时进行。

- 高难度和策略性（High Difficulty and Strategy）：这类游戏强调挑战性，玩家需要慎重决策，计划每一步行动。

- 资源管理（Resource Management）：玩家需要在有限资源的条件下生存，例如装备、生命值、魔法值等。

hack作名词，有“好的方案；好的建议”的含义，英文解释为“a good solution or piece of advice”如：a kitchen hack 可以理解为厨房窍门，厨房妙招，举个🌰：Here are 10 life hacks that will change your life for the better. 这是10个会让你人生更美好的建议。

作动词，表示“未经允许侵入他人的计算机系统（来窃取信息或从事非法勾当）”，英文解释为“to get into someone else's computer system without permission in order to find out information or do something illegal”举个🌰：Computer hacking has become widespread over the last decade. 计算机黑客现象在过去10年里非常普遍。

prominent /ˈprɒm.ɪ.nənt/ 1）表示“重要的；著名的；杰出的”，英文解释为“important or well known”举个🌰：He played a prominent part in the campaign. 他在这次运动中发挥了重要作用。

2）表示“突出的，显眼的”，英文解释为“Something that is in a prominent position can easily be seen.”举个🌰：New books are displayed in a prominent position on tables at the front of the store. 新书陈列在商店前面桌子上的一个显眼位置。

startup / start-up /ˈstɑːt.ʌp/ 表示“刚起步的小企业，新兴小型企业；初创公司”，英文解释为“a small business that has just been started”如：start-up costs 创业成本，举个🌰：Start-ups are very vulnerable in the business world. 商界中刚起步的小企业非常脆弱。

unprecedented /ʌnˈpres.ɪ.den.tɪd/ 表示“史无前例的，空前的；绝无仅有的”，英文解释为“never having happened or existed in the past”举个🌰：This century has witnessed environmental destruction on an unprecedented scale. 环境在本世纪遭到了空前的破坏。

词源和词根：un-是“不”的意思，precedented来自拉丁语的praecedere，意为“走在前面”。派生词：unprecedentedly，常用短语：unprecedented event 空前的事件。

🎬电影《星运里的错》（The Fault in Our Stars）中的台词提到：You are so busy being you that you have no idea how utterly unprecedented you are. 你太忙于做自己 都没有意识到自己是那么的令人心动。

The company behind ChatGPT said the startup Hugging Face had detected and contained the agent – an AI tool designed to carry out tasks without human assistance – which had entered its systems.

OpenAI公司（ChatGPT的开发商）表示，初创公司抱抱脸（Hugging Face）发现并控制了这个智能体——这是一种无需人类协助即可自主执行任务的AI工具——当时它已经悄然潜入该公司的系统。

1）表示“包含；容纳；盛”，英文解释为“to have something inside or include something as a part”举个🌰：How much liquid do you think this bottle contains? 你认为这个瓶可装多少液体？

2）表示“防止…蔓延（或恶化）”，英文解释为“to prevent sth harmful from spreading or getting worse”如：to contain an epidemic 防止流行病的蔓延。

3）表示“控制，克制，抑制（感情）”，英文解释为“to keep your feelings under control”举个🌰：She was unable to contain her excitement. 她无法抑制内心的激动。

🎬电影《哥斯拉》（Godzilla）中的台词提到：That's why our mission was to contain it. 所以我们的主要任务是控制它。

📍类似的一个词，restrain表示“阻止，制止；遏制”（to control the actions or behaviour of someone by force, especially in order to stop them from doing something, or to limit the growth or force of something）举个🌰：She was so angry that she could hardly restrain herself. 她气得几乎都控制不住自己。

📍类似的短语：keep/ get sth under control，举个🌰：You should learn to keep your temper under control. 你应该学会控制自己的脾气。

“We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities,” OpenAI said.

OpenAI公司表示：“我们认为这是一起史无前例的网络事件，其背后涉及最顶尖的网络攻击能力。”

state-of-the-art /ˌsteɪt.əv.ðiːˈɑːt/ 表示“最先进的；应用最新理念（或方法）的”，英文解释为“very modern and using the most recent ideas and methods”如：a state-of-the-art computer 最先进的计算机。

几个类似表达：

📍cutting-edge表示“尖端的，前沿的”，英文解释为“Cutting-edge techniques or equipment are the most advanced that there are in a particular field.”举个🌰：What we are planning is cutting-edge technology never seen in Australia before. 我们正在筹划的是澳大利亚前所未见的尖端技术。

🎬电影《像素大战》（Pixels）中的台词提到：I'm thinking this is a cutting-edge multinational enterprise. 我想是某先进跨国组织干的。

📍 leading edge表示“（某活动领域的）最重要位置，领先地位；（尤指技术上的）前沿，尖端”，英文解释为“The leading edge of a particular area of research or development is the area of it that seems most advanced or sophisticated.”如：at the leading edge of scientific research 在科学研究的前沿。

The company said it expected this type of incident to become more commonplace as models – the technology that underpins AI tools such as chatbots and agents – become more capable.

该公司预计，随着模型——即支撑聊天机器人和AI智能体等工具的核心技术——能力日益增强，这类事件将变得司空见惯。

表示“平常的；常见的；普通的；普遍的”，英文解释为“happening often or often seen or experienced and so not considered to be special”举个🌰：Electric cars are increasingly commonplace. 电动汽车越来越普及了。

🎬电影《偷书贼》（The Book Thief）中的台词提到：Unfortunately, I'm far more ordinary and commonplace. 不幸的是 我其实非常平凡。

underpin /ˌʌn.dəˈpɪn/ 表示“加强，巩固，构成（…的基础等）”，英文解释为“to support or form the basis of an argument, a claim, etc.”举个🌰：The report is underpinned by extensive research. 这份报告以广泛的研究为基础。

OpenAI said the hack occurred via an agent powered by a combination of its latest publicly available model, called GPT-5.6 Sol, and an even more capable model that is yet to be released.

OpenAI公司表示，这次入侵是由一个AI智能体发起的，该智能体融合了其最新公开模型GPT-5.6 Sol和一个尚未发布的更强大模型。

While being tested internally on their hacking capabilities in an enclosed digital laboratory known as a sandbox, the models gained open internet access – effectively an escape route – by locating a vulnerability that had not been discovered before.

在被称为沙盒（sandbox）的封闭数字实验室中，这些模型正在接受内部黑客能力测试。它们通过发现一个此前从未被察觉的漏洞，成功获得了开放互联网的访问权限——这无异于为自己开辟了一条“逃生通道”。

Sandbox（沙盒/沙箱）在网络安全与软件测试领域，指一个隔离的、安全的虚拟执行环境，它最核心的比喻是“儿童游乐场里的沙坑”——你可以在里面随便挖、堆、倒水，但玩完之后，沙子还是沙子，不会破坏外面的草地。

vulnerability /ˌvʌlnərəˈbɪləti/ 表示“易受伤害；有弱点；脆弱”，英文解释为“the quality of being vulnerable (= able to be easily hurt, influenced, or attacked), or something that is vulnerable”

The agent then hacked Hugging Face, which is a database of AI models, to locate technology that would help it pass the hacking evaluation. OpenAI said the models “successfully found ways to gain access to secret information that it could use to cheat the evaluation”. The attack ended when Hugging Face's security team and its own AI agents spotted and stopped the rogue activity.

随后，这个AI智能体入侵了抱抱脸——一个AI模型数据库——试图寻找能帮助它通过黑客评估的技术。OpenAI公司表示，这些模型“成功找到了获取秘密信息的方法，并利用这些信息来欺骗评估”。直到抱抱脸的安全团队及其自身的AI智能体发现并制止了这一恶意行为，攻击才宣告结束。

Hugging Face's chief executive, Clément Delangue, said the attack was “mind-blowing” but believed there was “no malicious intent” from OpenAI.

抱抱脸首席执行官克莱门特·德朗格（Clément Delangue）表示，这次攻击“令人瞠目结舌”，但他相信OpenAI公司并无“恶意”。

mind-blowing /ˈmaɪndˌbləʊ.ɪŋ/ 表示“令人印象深刻的；令人惊奇的”，英文解释为“extremely exciting or surprising”举个🌰：The special effects in this film are pretty mind-blowing. 这部电影的特技效果令人惊奇。

malicious /məˈlɪʃ.əs/ 1）表示“恶意的，恶毒的，意在伤人的”，英文解释为“intended to harm or upset other people”如：malicious gossip 用意歹毒的流言蜚语。

2）表示“（对计算机系统进行破坏或盗取个人信息）恶意的”，英文解释为“intended to cause damage to a computer system, or to steal private information from a computer system”如：protection against malicious software/code 保护不受恶意软件/代码伤害。

“We suspected last week's cyber-attack might have come from a frontier lab, given the sophistication of the agent,” he wrote on X.

他在X平台上写道：“考虑到这个AI智能体的复杂程度，我们怀疑上周的网络攻击可能来自某个前沿实验室。”

The term for an unknown IT flaw is a zero-day vulnerability because developers have zero minutes to fix the problem. In April, OpenAI's close rival Anthropic said its Mythos model had found thousands of these flaws.

未知的IT漏洞被称为零日漏洞（zero-day vulnerability），因为开发者根本没有时间来修复它。今年4月，OpenAI公司的劲敌Anthropic公司表示，其Mythos模型发现了数千个此类漏洞。

rival /ˈraɪ.vəl/ 作名词，表示“竞争对手；敌手”，英文解释为“a person, group, etc. competing with others for the same thing or in the same area”

作动词，表示“可与…相媲美，比得上”，英文解释为“to be as good, clever, beautiful, etc. as someone or something else”举个🌰：No computer can rival a human brain for/in complexity. 没有任何计算机比得上人脑复杂。

The revelation of Mythos's ability to locate and exploit zero days led to the US government restricting exports of Mythos and its sister model Fable 5, although it has since lifted the ban. GPT-5.6 Sol had similar restrictions but has since been rolled out worldwide.

由于Mythos模型能够定位并利用零日漏洞，美国政府曾限制其及其姊妹模型Fable 5的出口，不过后来解除了禁令。GPT-5.6 Sol也曾受到类似限制，但此后已在全球范围内推出。

Greg Casar, a Democratic US congressman, said the incident was alarming.

美国民主党众议员格雷格·卡萨尔（Greg Casar）表示，这起事件令人担忧。

“AI is developing extremely fast with no real regulations to keep us safe,” he said in a statement calling for mandatory independent safety testing, mandatory disclosure of security incidents and international cooperation “to keep people safe from absolute disaster”.

“人工智能发展极其迅速，却没有真正的法规来保障我们的安全，”他在一份声明中说，并呼吁进行强制性独立安全测试、强制性披露安全事件以及国际合作，“以保护人们免遭灭顶之灾”。

mandatory /ˈmæn.də.tər.i/ 表示“强制的；必须履行的；法定的”，英文解释为“Something that is mandatory must be done, or is demanded by law.”举个🌰：Athletes must undergo a mandatory drugs test before competing in the championship. 运动员在锦标赛前必须进行强制性药检。

disclosure表示“透露的信息；公开，透露，披露”，英文解释为“a secret that someone tells people, or the act of telling this secret”

- 词汇盘点 -

- 词汇助记 By AI -

LearnAndRecord

2015年2月8日

2026年7月23日

第4184天

每天持续行动学外语