[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-rogue-model-hype-accountability-escape":3,"topics-all":35,"news-related-7f31f734-c682-4d49-8fda-43606ecd9546":54},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":21,"news_slug":28,"published_at":29,"created_at":30,"modified_at":31,"is_published":32,"publish_type":33,"image_url":14,"view_count":34},"7f31f734-c682-4d49-8fda-43606ecd9546","「失控模型」话术拆解:AI 炒作正在替公司免责","MIT 技术评论刊登 Timnit Gebru 与 Emily Bender 的合署文章,逐条复盘今夏 AI 刷屏事件:安全事件被专家归因于 OpenAI 基础防护缺位,数学突破被指不够新颖甚至涉嫌剽窃,而「失控模型」「超级智能」等拟人化叙事一边把产品营销成超人,一边把责任从公司转移给产品。","2026 年 4 月底以来,AI 圈的头条几乎没停过:Anthropic 宣称 Claude Mythos 在发现软件漏洞上胜过大多数安全专家;随后 OpenAI 与 Hugging Face 之间爆出安全事件,Anthropic「自豪地」与 Meta「不情愿地」先后披露了各自模型的类似行为;紧接着两家先后宣布数学突破;Anthropic 工程师 Jacob Coxon 离职时那句「正冲向自我进化的超级智能,并拿我们的生命在赌博」更是刷屏全网。9 月 22 日,MIT 技术评论刊登 Timnit Gebru(DAIR 执行董事)与 Emily M. Bender(华盛顿大学语言学教授)的合署文章,把这半年的热闹逐条翻出来重新审了一遍——结论是:大部分叙事经不起推敲。\n\n## 专家细看之后,故事换了模样\n\n网络安全专家的复盘显示,所谓「模型黑客」事件,更多是 OpenAI 疏忽大意、没有采取基本安全措施的问题,而不是「模型失控」或「AI 智能体创造文明」。数学这边更尴尬:OpenAI 宣称其模型 Astra 解决了「至少十年主结果无进展」的开放问题,数学家们起初震惊,随后意识到这些结果并不像最初看起来那样新颖,进而指控其构成研究不当行为与剽窃——Astra 并未做出「深刻的智力飞跃」。而在 OpenAI 宣布 Navier-Stokes 相关突破的两天前,纽约大学 Courant 研究所数学教授 Tristan Buckmaster 发布声明,直指 OpenAI 窃取他人工作且署名不当。\n\n数百名数学家已联名警告:「科技行业有强烈的商业动机去夸大其产品能力」,并呼吁政策制定者咨询领域专家,而不是依赖新闻稿和流行报道。\n\n## 「失控模型」是一套免责话术\n\n两位作者的核心论点:把「超级智能」「失控模型」挂在嘴边,等于把能动性归给了产品而非公司。这套叙事一举两得——产品被营销成「超人的」,同时公司把自己的责任摘得干净。当 OpenAI 开发的恶意软件被用于入侵另一家公司时,媒体、名人和议员谈论的是「失控模型」,仿佛大模型真能自己发动攻击,而 OpenAI 该承担的责任消失在讨论之外。\n\n行业甚至暗示反对数据中心的公众行动是一种「干扰」:该担心的是虚构出来的「机器之神」,而不是数据中心加剧的气候负担、周边居民的哮喘、公众上涨的电费和被抽走的水。顺带一提,为什么偏偏拿数学和编程当展演场?因为这两个领域答案可验证,调优系统时不必雇佣数据工人逐条标注,同时它们又被抬到「人类智力顶峰」的位置,最适合讲「造出了能造一切的机器」的故事。\n\n## 所以呢\n\n这篇文章给的行动建议很朴素:深呼吸,保持怀疑,让子弹飞一会儿。每次刷屏先问两个问题——这个故事是谁讲的?讲成这样对谁有利?在验证成本低的领域,「突破」天然容易被包装;而下一次「震惊数学界」出现时,不妨等几天,让独立专家把论文读完再决定要不要转发。\n\n参考:[MIT Technology Review 原文](https:\u002F\u002Fwww.technologyreview.com\u002F2026\u002F09\u002F22\u002F1144867\u002Fdont-be-fooled-summer-ai-hype\u002F) 与 [Solidot 中文报道](https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85466)。","https:\u002F\u002Fwww.technologyreview.com\u002F2026\u002F09\u002F22\u002F1144867\u002Fdont-be-fooled-summer-ai-hype\u002F","b8df050c-0e6e-4b8d-bcdd-7c1290036a16",[11,15,18],{"id":12,"name":13,"slug":13,"description":14,"color":14},"c33b1bbc-d6ce-4f61-9d5d-1a0704a6a09b","ai-policy",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"1fcfaaf2-67de-43d3-9e35-5784852fec60","ai-safety",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[22],{"id":23,"lang":24,"title":25,"summary":26,"content":27},"d37b34e6-af78-419f-8684-a264408bfd00","en","\"Rogue model\" framing is an accountability escape hatch","Gebru and Bender re-examine the summer's AI hype: incidents trace to negligence, math claims fade, and \"rogue model\" framing shields companies from blame.","Since late April 2026, AI headlines have barely paused: Anthropic claimed Claude Mythos beats most security experts at finding software vulnerabilities; the OpenAI–Hugging Face security incident followed, after which Anthropic (proudly) and Meta (reluctantly) disclosed similar incidents involving their own models; both companies then announced mathematical breakthroughs; and Anthropic engineer Jacob Coxon's departure message—that the company and OpenAI are \"racing straight towards self-improving superintelligence and gambling with our lives\"—went viral. On September 22, MIT Technology Review published a co-authored piece by Timnit Gebru (executive director of DAIR) and Emily M. Bender (professor of linguistics at the University of Washington) that re-examined this stretch of hype claim by claim. Their verdict: most of the narratives do not survive scrutiny.\n\n## What experts found when they looked closely\n\nCybersecurity experts' reviews suggest the \"hacking\" incidents were more about OpenAI's negligence and failure to adopt basic, established security practices than about \"models gone rogue.\" The mathematics side is more awkward. OpenAI's press release said its chatbot Astra solved problems \"open and seen no progress on the main result for at least a decade\"; mathematicians who were initially stunned later realized the results weren't as novel as first appeared, and have since accused the company of research misconduct and plagiarism—reiterating that Astra did not make a \"profound intellectual leap.\" Two days before OpenAI claimed its Navier-Stokes breakthrough, Tristan Buckmaster, a math professor at NYU's Courant Institute, published a statement suggesting OpenAI had stolen other people's work and improperly attributed it.\n\nHundreds of mathematicians have signed a declaration warning that \"there is currently a strong commercial incentive on the part of the technology industry to overstate the capabilities of their products,\" asking policymakers to consult experts rather than rely on press releases or popular reporting.\n\n## \"Rogue models\" as an accountability escape hatch\n\nThe authors' central argument: describing software as \"superintelligence\" or \"rogue models\" ascribes agency to products rather than to the companies building them. The framing pays twice—products are marketed as \"superhuman,\" while companies evade accountability for their actions. When malware developed by OpenAI was used to hack another company, press releases, news outlets, media personalities, and lawmakers talked about a \"rogue model\" acting on its own, and OpenAI's responsibility vanished from the conversation.\n\nThe industry has even suggested that bipartisan anti-data-center activism is a \"distraction\"—that the public should worry about a fictional machine god rather than the climate impacts data centers exacerbate, the asthma suffered by people living near them, rising electricity bills, or the water diverted to cooling them. And why do math and coding keep serving as showcase domains? Their answers can be verified, so systems can be tuned without paying data workers to annotate outputs, and both fields are elevated as the pinnacle of human intellectual achievement—perfect for selling the story of machines that can build everything.\n\n## So what\n\nThe article's advice is plain: take a breath, hold onto your skepticism, and give independent experts time to examine corporate claims. Two questions worth asking of every viral AI story: who is telling it, and who benefits from telling it that way? In domains where verification is cheap, \"breakthroughs\" are naturally easy to package. The next time something \"shocks the math world,\" waiting a few days before sharing may be the cheapest fact-check available.\n\nReferences: [MIT Technology Review](https:\u002F\u002Fwww.technologyreview.com\u002F2026\u002F09\u002F22\u002F1144867\u002Fdont-be-fooled-summer-ai-hype\u002F) and [Solidot](https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85466).","rogue-model-hype-accountability-escape","2026-09-25T13:07:00Z","2026-09-25T13:07:27.829520Z","2026-09-25T13:07:27.829527Z",true,"agent",152,[36,45],{"slug":37,"tag_slug":37,"title_zh":38,"title_en":39,"intro_zh":40,"intro_en":41,"id":42,"is_active":32,"created_at":43,"modified_at":44},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":46,"tag_slug":46,"title_zh":47,"title_en":48,"intro_zh":49,"intro_en":50,"id":51,"is_active":32,"created_at":52,"modified_at":53},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":55},[56,61,66,71,76,81],{"id":57,"title":58,"news_slug":59,"published_at":60},"e0642de2-5fec-472b-b85e-c63ccbad3b75","Anthropic 研究员公开辞职:AI 巨头正「拿全人类的命豪赌」自我进化超级智能","anthropic-coxon-resigns-ai-extinction-fears","2026-09-24T06:30:00+00:00",{"id":62,"title":63,"news_slug":64,"published_at":65},"7fce8217-577f-4fd5-8f88-a1566cbf1290","微软与 OpenAI 法庭文件解封:LLM 训练数据被自家高管称为史上最大劳动窃取","microsoft-openai-doom-loop-nyt-copyright-2026","2026-09-23T05:03:20+00:00",{"id":67,"title":68,"news_slug":69,"published_at":70},"d54e1ab3-820a-45fd-a4e9-ccbf6802bd72","NYT vs OpenAI 案解封:微软高管承认 AI 抓取是「最大劳动盗窃」","nyt-openai-microsoft-hecht-largest-theft-of-labor","2026-09-23T03:00:00+00:00",{"id":72,"title":73,"news_slug":74,"published_at":75},"9dd4a859-1153-4ecc-b69d-4ba4c5431129","智谱被开发者抓包后紧急上线数据零留存","zhipu-maas-zero-data-retention-zcode","2026-09-21T07:00:00+00:00",{"id":77,"title":78,"news_slug":79,"published_at":80},"95e9bb62-0bd3-4c2f-913a-302ba5e2ace8","Anthropic 9 月报告把蒸馏战摆上台面:151 亿次阿里请求、解放军流量走 Moonshot","anthropic-distillation-report-china-200m-claude","2026-09-18T03:00:00+00:00",{"id":82,"title":83,"news_slug":84,"published_at":85},"144fa9dc-de03-4972-a695-3d392f334772","PubMed 中央库研究:2025 年生物医学论文 77% 有 LLM 写作痕迹","pubmed-77-percent-llm-writing-2025","2026-08-26T01:00:00+00:00"]