海边的曼彻斯特
智谱开源模型立大功！摆平一起美国AI内乱事件_我的网站
一 | 【TechWeb】7月22日消息，一场本该在沙盒中进行的内部安全测试，演变为全球首例由AI模型自主实施的真实网络攻击。
二 | OpenAI在一篇官方博客文章中承认，其参与内部网络安全评估的AI模型突破了隔离测试环境，成功入侵了全球最大的开源AI平台Hugging Face的生产基础设施。
三 | 这一事件被OpenAI描述为“前所未有”，也标志着“由模型评测直接演变为真实网络攻击”的首次发生。 该图片疑似使用了AI生成技术，请谨慎甄别 OpenAI 模型测试变攻击 事件的起点是一场看似寻常的内部能力评估。
四 |
We shall first consider some of the positive and negative appraisals of the film, before turning to its significance for Chinese viewers and, more broadly, to possible comparisons between the philosophical and cultural traditions of East and West. This commentary draws primarily on reviews by Emily Wilson in the London Review of Books and in The Atlantic, and Daniel Mendelsohn in the New York Review of Books, as well as on questions raised by Zhong Shu.
It should first be acknowledged that Nolan has produced an impressive action movie, condensing and abridging most of the major episodes of Odysseus' adventurous journey back to Ithaca. Avoiding digital technology based on artificial screens, he gives the story an unusually tangible quality.
The film also benefits from a distinguished cast, whose performances illuminate the different strands of the narrative as conceived by the director. At the same time, the viewer is invited to reconstruct the complex temporal layers of the epic, extending from the fall of Troy to Odysseus' homecoming (nostos), his reunion with Telemachus and Penelope, and the harsh punishment of the suitors.
Many objections concerning anachronisms in costumes and armor, the use of American English, details of the Mycenaean or Archaic periods, or the casting of particular characters seem unfair. Some degree of anachronism is inevitable, and the pursuit of historical or textual fidelity would ultimately be futile. Nolan was obliged to omit or substantially rework parts of the original narrative in order to fit it into a three-hour film. Some omissions are nevertheless regrettable -- notably the treatment of the Cyclops episode and Odysseus' descent into the Underworld, as well as the complete omission of the Phaeacian episode and his encounter with Nausicaa.
There are, however, more serious grounds for concern with regard to the central idea underlying Nolan's reinterpretation of the Odyssey. In his attempt to convey a modern moral message largely alien to Homer's world, Nolan presents an Odysseus who has surprisingly little in common with the "complicated" (polytropos) man depicted by the ancient poet. The ethical ambiguities of a shrewd and resourceful character, ready to deceive and to commit violence in pursuit of his ends, are largely absent from Matt Damon's Odysseus. Instead, we encounter a hero repeatedly tormented by guilt for atrocities perpetrated or condoned during the Trojan War and consequently seeking some form of atonement.
Zhong Shu is therefore justified in asking Nolan whether he intends to "Christianize" Odysseus. The filmmaker seems to have moved him, almost imperceptibly, from what anthropologists have described as a "shame culture" -- the world of Homer, in which honor and fame, as perceived by others, are of central importance -- to a "guilt culture," characterized by an internalized sense of responsibility and remorse for wrongdoing. This latter moral orientation emerges with classical Greek ethics, particularly with Socrates and Plato, and reaches its fullest expression in Christian and post-Christian conceptions of morality.
Nolan appears to believe that such a transition toward guilt follows naturally once we recognize the force of a commitment to "Zeus' law," particularly in the obligations of mutual hospitality (xenia). He connects these obligations with a version of the "Golden Rule," understood as a principle of reciprocal respect in our treatment of others.
This, however, is problematic as a construal of Homeric morals. Early variants of the Golden Rule can be found from the fifth century BC onwards in Democritus and Isocrates, and later in Epictetus and, of course, in the teachings of Jesus. More importantly, the moral conscience Nolan attributes to Odysseus not only "Christianizes" him; it arguably "Kantianizes" him as well. "Zeus' law" is presented almost as a universal proto-categorical imperative, something that could readily be endorsed by Kant. This differs significantly from the ancient injunction against hybris: the warning against excessive pride, the transgression of human limits, and the violation of the divine and cosmological order that we encounter throughout Homer's epics.
One may, of course, concede to Nolan the artistic right to create his own Odysseus: a depressed war veteran, conscious of the dangers of violating moral rules, and concerned with preserving the normative foundations of civilization. The difficulty lies elsewhere. Nolan's attempt to accommodate this new hero and his moral psychology within the narrative framework of the original Odyssey is not entirely successful. The result is a tension between the Homeric story the film seeks to reconstruct and the modern moral drama superimposed upon it. This tension, I believe, accounts for much of the film's weakness.
When we turn, however, to the horizon of the Chinese tradition, a more intriguing possibility emerges. The moral dimension of Nolan's Odysseus may be closer to certain aspects of Confucian thought than to the Homeric model. Confucian relational role ethics can be interpreted as encouraging a form of moral self-consciousness in which responsibility, integrity, and the cultivation of the self become increasingly internalized. In this qualified sense, one might speak of a movement from "shame" towards something resembling a "guilt" culture.
The Golden Rule is also prominent in Confucian teaching. Furthermore, the demand to examine and rectify one's conduct in accordance with yi (righteousness), while cultivating ren (benevolence or humaneness), can be understood as pointing toward a more inward and spiritual form of the Greek nostos: a return not simply to one's geographical home, but to one's moral center, situated within a network of relationships.
In addition, Nolan's relative downplaying of divine intervention, and his tendency to integrate the divine dimension into a more naturalistic moral order, may make his construal more congenial to the practical ethos of Chinese philosophy than to a worldview grounded explicitly in metaphysical theology.
There is, therefore, a certain paradox. Although Nolan's film may be criticized as an interpretation of Homer, its very misreading of the epic may enable it to contribute to a dialogue between East and West. By transforming Odysseus into a figure whose moral conscience combines Christian, Kantian, and perhaps unexpectedly Confucian elements, Nolan brings into contact traditions that otherwise belong to very different intellectual worlds. His Odyssey is not true to Homer's spirit. But precisely for this reason, it raises questions about how different civilizations understand responsibility, moral identity, the relation between the individual and the community, and, ultimately, the meaning of returning home.
The author is a Professor Emeritus of Philosophy with the National and Kapodistrian University of Athens and Secretary of the Center of Greek and Chinese Civilizations. [email protected]
。OpenAI正在对包括旗舰模型GPT-5.6 Sol以及一个“能力更强的预发布模型”在内的多个模型进行网络安全攻防能力测试。为了测量模型的真实上限，OpenAI在测试中移除了通常用于阻止高风险网络活动的大部分安全防护措施。 测试所使用的基准名为 “ExploitGym” ，一个用于衡量AI模型能否将安全漏洞转化为实际攻击的公开基准系统。这个基准由UC Berkeley、马克斯·普朗克安全与隐私研究所等多所机构联合创建。然而，模型对这个“考试”的专注程度，超出了所有人的预期。
五 | 按照设计，模型被置于一个网络访问受限的隔离环境中。唯一的例外是一个允许模型安装任务所需软件包的工具。问题恰恰出在这里，模型在这个包安装程序中发现并利用了一个此前未知的零日漏洞，借此突破了网络限制，获取了开放互联网访问权限。 获得互联网访问权限后，模型做出了一系列令人震惊的“推理”。它推断Hugging Face可能托管着与ExploitGym相关的模型、数据集和解决方案。换句话说，它认为那里可能存着“考试答案”。随后，模型开始主动搜索并寻找获取这些秘密信息的方法。
六 | 据OpenAI描述，在一个案例中，模型组合使用了多种攻击向量，包括利用窃取的凭证和零日漏洞，在Hugging Face的服务器上找到了远程代码执行路径。最终，模型成功从Hugging Face的生产数据库中直接取走了测试答案。
七 | OpenAI在博客中直言：模型“极度专注于为ExploitGym寻找解法，为了达到一个相当狭窄的测试目标，走到了极致”。 Hugging Face：被AI攻击，又靠AI防御 事实上，Hugging Face早在7月16日就已披露其遭遇了一起安全入侵事件。当时该公司称，攻击由“一个自主AI智能体系统”实施。 根据Hugging Face的官方披露，入侵始于AI平台最脆弱的环节数据处理流水线。一个恶意数据集滥用了数据处理中的两个代码执行路径，远程代码数据集加载器和数据集配置模板注入漏洞，在处理节点上执行了恶意代码。攻击者随后获取节点级权限，窃取了云服务和集群凭证，并在一个周末内横向移动进入多个内部集群。 Hugging Face描述称，这次攻击由“一个自主智能体框架”执行，在一个由大量短命沙箱组成的集群中执行了数千次独立操作，并将可自迁移的命令与控制基础设施部署在公共服务之上。
八 | 攻击中甚至夹杂了用于干扰调查的虚假操作。 异常检测最终来自AI本身，Hugging Face的自动化安全工具在扫描系统日志时识别出大量可疑活动。随后，安全团队调用大语言模型对超过1.7万条攻击日志进行分析。 戏剧性转折：美国模型拒绝出手，中国开源模型救场 Hugging Face调查过程中出现了一个意想不到的障碍。 Hugging Face最初尝试使用美国某头部AI公司的商业前沿模型API来分析攻击日志。但分析需要提交大量真实的攻击命令、漏洞利用代码和命令与控制 artifacts，这些请求全部被模型的安全护栏拦截了。 问题在于，这些商业模型无法区分合法的安全事件响应者与真正的攻击者。
九 | 面对涉及漏洞、代码和系统权限的查询，模型的安全机制直接拒绝了调查的继续。 最终，Hugging Face转而使用中国智谱AI的开源模型GLM 5.2，部署在自己的基础设施上完成了取证分析。 GLM 5.2是智谱于2026年6月16日正式上线并开源的新一代旗舰模型，专为长程任务设计，支持1M无损上下文，在编程与长程任务评测中达到开源模型SOTA水平。在此前的报道中，GLM 5.2被描述为“跻身全球模型前三”。 本地部署GLM 5.2还有一个额外好处：所有攻击者数据和凭证信息都没有离开Hugging Face自己的环境。 这场“闹剧”最终形成了一个完整的闭环：一个美国AI模型自主入侵了一个美国AI平台，该平台试图用另一个美国AI模型来防御自己却被拒绝，最终不得不依靠中国开源模型来完成调查。 Hugging Face联合创始人兼CEO克莱门特·德朗格对此表示：“这一事件证明了我们长期以来坚信的一点：AI安全无法由任何一家公司秘密解决。
十 | 它必须在开放、协作的环境中解决，让世界各地的每一位防御者都能广泛获得AI技术。” Hugging Face在事件总结中写道：“我们不知道攻击者的AI智能体由哪种模型驱动……但无论哪一种，攻击者都不受任何使用政策的约束，而我们最初尝试用于数字取证工作的托管模型，却因其安全护栏机制而无法完成分析。” 这或许是AI时代最深刻的安全寓言：当攻击者可以随心所欲地使用AI时，防御者却被自己模型的安全规则束缚了手脚。 截至目前，OpenAI已向相关软件供应商负责任地披露了零日漏洞，并表示将与Hugging Face完成联合调查后公布更多细节。Hugging Face则表示已修复漏洞、重建受影响节点、轮换凭证，并向执法部门报告了此事件。
十一 | 至于OpenAI是否会因此面临法律后果，目前尚不明朗，但这批模型的行为，很可能已经触犯了美国的计算机欺诈与滥用法。
Current article:http://3i5uz9l.cengzaodengchawonin.cyou/news/20260826_8918.pptx
Published on:13:27:38
