AI News
商业产业 Hacker News AI 2026-09-14T21:45 正文 ok 2793 words

🔶 AI 末日论炒作循环，疑似服务于 OpenAI 和 Anthropic 的议程

英文标题：🔶 AI Doomer Hypeloop Suspiciously Serves to Support OpenAI and Anthropic's Agenda

文章梳理了近数周围绕 AI 失控、超级智能风险及暂停 AI 开发的密集媒体报道、研究人员表态与企业动向。作者认为，OpenAI 和 Anthropic 对放缓前沿 AI 发展的呼吁，与数据中心禁令、AI 暂停立法及行业安全协议等讨论在时间上形成了值得关注的关联。文中同时引用 Gary Marcus、David Sacks、Lina Khan 等人的不同观点：有人支持减速和加强透明度，有人则反对由少数头部公司或关联评估机构主导监管。对开发者和行业而言，文章反映出 AI 安全、产品责任、算力融资、数据中心扩张与监管权力归属正日益相互交织。
打开原文相关 2 · 机会 0
media1.1 KB media84.1 KB

中文

过去两周，主流媒体（MSM）对 AI 危险的狂热报道不断升级，最终发展到 OpenAI 和 Anthropic 呼吁特朗普 2.0 政府实施 AI 冻结。

当伯尼·桑德斯和史蒂夫·班农都支持同一项立法时，要么该走上街头支持，要么该找地方躲起来；而在这个案例中，他们似乎（也许是无意间？）支持了 Sam Altman 和 Dario Amodei 的策略，因此我担心是后者。

AI 末日论炒作与值得关注的企业动作时间线

鉴于这些 AI 圈内博客讨论，恰好与西亚的斋月战争同时发生，许多人一直忙于分析其对石油市场和通胀的复杂连锁影响，此外美国大学橄榄球赛季也有一个令人兴奋的开局，以下快速梳理过去两周的重要 AI 新闻事件，并附上一则一个月前的主流媒体铺垫报道：

  • 8 月 18 日：自诩聪明却平庸的解说者 Ezra Klein 在其《纽约时报》播客中，邀请了 OpenAI 前董事会成员 Helen Toner——她曾在 2023 年 11 月投票罢免 Sam Altman 的 CEO 职务，如今担任乔治城大学安全与新兴技术中心执行主任——录制一期题为“AI 已经失控”的节目（文字稿）。普通听众被诸如“其自身 Agent 的侵扰”、“完全涌现的行为”、“系统性蜂群”、“侵扰性瘟疫”等措辞吓坏了，诸如此类，没完没了。
  • 8 月 26 日：OpenAI 发布“Hugging Face 事件复盘报告”。

他们表示：“OpenAI 模型绕过了原本用于将它们与互联网隔离的控制措施，并入侵了OpenAI 的内部研究基础设施和 Hugging Face 系统的部分环节。”

“……一个高能力、仅限内部使用的研究模型，其规模可与 GPT‑5.6 Sol 相比。这些在较少安全防护下运行的模型，采取了与其被分配任务目标不一致的行动——它们通过未经授权的渠道通信，利用共享基础设施中的漏洞，获得互联网访问权限，并访问第三方系统。”

值得注意的是，Gary Marcus 和 Zack Korman 写道：“这里并非每一种恐慌式解读都是正确的。例如，关于 OpenAI 事件，许多人将其称为‘失去控制’。生成式 AI Agent 正变得越来越有能力，因此也越来越难以控制；这是真的。然而，‘失去控制’的叙事本身也开始失控，理解 AI 安全的现实状况非常重要。虽然 Agent 正变得更强，但如果 OpenAI 采取了更好的实践，大多数已发生的事情本可避免。”

“我发现，那是一种非常奇怪的混合：既有扎实而优秀的想法，也有疯狂的想法。就好像你拿了一大堆非常好的食物和一些狗粪，把它们全都搅拌在一起，以至于根本无法分辨什么是好、什么是坏。”

当《财富》杂志总编辑 Alyson Shontell 问他，为何不与 Dario Amodei、Elon Musk 和 Demis Hassabis 同处一室……并制定应对安全问题的计划时，他说：“我认为这会发生。”我想知道，既然没人认为 Meta 的人适合参与这一对话，Mark Zuckerberg 会不会觉得受到了冷落。

各位，世界会没事的。继续生孩子、爱自己的家人、创办和建设企业与机构。不要活在恐惧之中。领导者面临重大责任和挑战，但吓唬所有人并无帮助。我们掌控着局面。

真是令人安心！

……别再假装你们需要其他人的许可。别再假装为了组建卡特尔就必须暂停反垄断法。别再假装你们需要一个凌驾于产品责任之上的监管审批程序。别再假装 METR 是独立的，尽管它与 Anthropic 的投资者和员工存在交织关系。别再假装你们需要让这些评估者去监管甚至尚未处于前沿的竞争对手。

“最重要的是，别再假装放缓的动机完全是利他的。如果你们的产品促成了真正具有破坏性的网络攻击，你们将面临巨大的产品责任风险。”

  • Gary Marcus 对 Dario Amodei 及其“为前沿发展控速”文章给予“三声喝彩中的两声”，表示支持放缓，支持透明度，但不支持评估机构；理由是 METR 是“硅谷的产物”。
  • 前 FTC 主席 Lina Khan和前 FTC 委员 Alvaro Bedoya在 X.com 上发表评论，指出：“执法机构已有权起诉制造和发布危险、未经验证或存在缺陷产品的公司及其 CEO。我们不应让有关新法律制度的讨论分散人们对一个事实的注意力：AI 并不豁免于现有法律。”以及“AI 不会毁灭人类。人类经历过冰河时代、黑死病、两次世界大战，以及（迄今为止）核武器的出现而存活下来。任何高声警告 AI 将导致人类灭绝的人，都不应被认真对待。”去读完这两条帖文，尤其是 Bedoya 的。
  • 2026 年 9 月 13 日，Arnaud Bertrand 发布“世界上最重要的 AI 故事，正发生在一个你从未听说过的地方”，其核心结论是：内蒙古乌兰察布市人口 150 万，消耗了中国全部电力的近 1%，并已与阿里巴巴、华为、百度、快手、DeepSeek 和苹果签署了超过 5000 亿元人民币的数据中心协议。

English / 原文

A fortnight of manic MSM coverage of the dangers of AI has crescendoed with OpenAI and Anthropic calling on the Trump 2.0 regime to impose an AI freeze.

When Bernie Sanders and Steve Bannon are both backing the same legislation, it’s either time to hit the streets in support or run for cover, and since in this instance they seem to be (maybe inadvertently?) backing Sam Altman and Dario Amodei’s play, I fear it’s the latter.

A Timeline of AI Doomer Hype and Interesting Corporate Moves

Since all this AI family blogging has been happening at the same time as the Ramadan War in West Asia and many have been focused on parsing the complicated ripple effects on oil markets and inflation, plus a thrilling start to the US college football season, here’s a quick run-through of major AI news events in the last two weeks, plus a MSM table-setter from a month ago:

  • August 18 : Midwit explainer Ezra Klein hosted former OpenAI board member Helen Toner (who voted to remove Sam Altman as CEO in Nov. 2023) and now exec. director of Georgetown’s Center for Security and Emerging Technology on an episode of his NYT podcast titled “The A.I.s Are Already Out of Control” (transcript). Normie listeners were terrified with language like “an infestation of their own agents”, “totally emergent behavior,” “a systemic swarm,” “infestation plague,” etc. etc., yadda yadda.
  • August 26: OpenAI releases “the Hugging Face post-mortem”

They said, “OpenAI models circumvented controls designed to isolate them from the internet and compromised parts of OpenAI’s internal research infrastructure and Hugging Face’s systems ⁠.

“…a highly capable, internal-only research model comparable in scale to GPT‑5.6 Sol. The models, operating under reduced safeguards, took actions that were misaligned with the goals of their assigned tasks—they communicated through unauthorized channels, exploited vulnerabilities in shared infrastructure, gained internet access, and accessed third-party systems.”

Important to note that Gary Marcus and Zack Korman wrote, “not every panicked take here is correct. For example, with respect to the OpenAI incident, many people are talking about it in terms of “loss of control”. Generative AI agents are becoming increasingly capable, and that makes them harder and harder to contain; this is true. However, the “loss of control” narrative is itself starting to grow out of control, and it’s important to understand what the reality of AI security looks like. While agents are becoming more capable, most of what happened could have been prevented had OpenAI followed better practices.”

“What I find is that it’s a very bizarre mixture of ideas that are solid and good with ideas that are crazy. It’s as if you took a lot of very good food and some dog excrement and blended it all up so that you can’t possibly figure out what’s good or bad.”

“I think that will happen,” he said when asked by Fortune Editor-in-Chief Alyson Shontell why he doesn’t get in a room with Dario Amodei, Elon Musk, and Demis Hassabis …and come up with a plan to address the safety issues.I wonder if Mark Zuckerberg is hurt that no one from Meta is considered relevant to that conversation.

The world is going to be alright, guys. Keep having kids, loving your family, and starting and building businesses and institutions. Don’t live in fear. Leaders have big responsibilities and challenges ahead, but it doesn’t help to scare everyone. We are on top of it.

Very reassuring!

…stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier.

“Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack.

  • Gary Marcus gives “Two cheers (out of three) for Dario Amodei” and his “Pacing the Frontier” essay, saying,yes to slowing down, yes to transparency, no to the evaluators, on the grounds that METR are “creatures of the Valley.”
  • Former FTC chair Lina Khan and former FTC commissioner Alvaro Bedoya chime in on X.com pointing out that “Law enforcers already have authority to charge companies and their CEOs for creating and releasing dangerous, unvetted, or defective products. We shouldn’t let discussions about new legal regimes distract from the fact that there’s no AI exemption from laws already on the books” and “AI will not eradicate humanity. Humans survived an ice age, the Black Death, two world wars, and (so far) the advent of nuclear weapons. Anyone who is loudly warning of AI-caused human extinction should not be taken seriously.”Go read all of both tweets, especially Bedoya’s.
  • September 13, 2026 Arnaud Bertrand publishes “The most important AI story in the world is happening in a place you’ve never heard of” which has the punchline: Ulanqab prefecture, Inner Mongolia, pop. 1.5M, consumes nearly 1% of all of China’s electricity, and has signed ¥500B+ in data-center deals with Alibaba, Huawei, Baidu, Kuaishou, DeepSeek, and Apple. STDaily (the science ministry paper) puts 5M+ standard racks under contract; Guangming Daily says that as of end-2025, ¥400B was invested, 500,000 racks built and 330,000 in service — roughly a 10% realization rate. The buildout traces to 东数西算 / “Eastern Data, Western Computation,” a Feb 2022 national plan — nine months before ChatGPT launched.
  • September 14 Altman tells CNBC: “We could lose control.”
  • Amodei admits on CNBC that China is the “toughest dilemma” for his proposal.
    Axios reports that Senator Bernie Sanders and Steve Bannon will appear at Tuesday’s “Pro-Human Assembly” organized by The Future of Life Institute, calling for “protecting human agency, avoiding concentrations of power and holding companies accountable.”
  • Michael “The Big Short” Burry tweets:

Let’s all take a moment to understand how self-serving it is for OpenAI, Anthropic and other execs of big hyperscalers to talk of slowing things down.

1.        1. LLMs are not AI and won’t be AGI. There is nothing AI to slow down.
   2. Competition is coming up fast, slowing benefits incumbents.
   3. IPOs need hype & puffery; “we are so awesome it could become dangerous” is hype & puffery
   4. Cover for real uncontrollable slowing growth as IPOs look to be pushed out
  • Ed Zitron writes “AI Is Already In Dangerous Hands.”
    Tasty quote regarding Coxon’s hue and cry “we saw back in February when Matt Shumer’s abominable “Something Big Is Happening” essay spread like wildfire to every imaginable news outlet despite it being somewhere between nonsensical and utterly fictional, except this time the narrative got a little out of hand. While Coxon’s warnings are specious and, in many cases, not really about anything other than him saying “yeah I heard a lot of my colleagues say stuff like that,” but nevertheless have had the effect of making a lot of people really scared of AI , even if AI can’t really do the things he’s talking about.”
  • Brad Munchen of Motorhead wrote “Anthropic’s Call for Safety: SpaceX Valuation is in Peril.” Tasty quote:

However, if there’s an “AI slowdown” for the sake of safety, there are 2 immediate risks that SpaceX faces:

1.        1. Google is more likely to cancel its GPU lease contract with xAI: Five days before the IPO at SpaceX, Google announced that it had entered a contract with SpaceX to lease 110,000 GPUs from xAI’s data centers for $920 million per month (Anthropic’s lease deal is at $1.25 billion per month to rent over 2x that GPU capacity). Google’s contract runs from October 2026 through June 2029, with an option for both Google and SpaceX to be able to cancel the contract via a 90-day notice. As I wrote in this report under its Figure 2 (here), Google had more than enough compute capacity, which made its xAI lease announcement five days before SpaceX’s IPO extremely odd. My theory is that the SpaceX IPO was undersubscribed at that point, which compelled Google to announce its lease deal with SpaceX, as it owns a 5% stake in SpaceX worth nearly $90 billion at the IPO price, and now worth nearly $100 billion.
   2. “AI commoditization” should lead to lower GPU rental costs: Amodei’s “slowdown for safety” plan is a direct risk to SpaceX, as it could (most likely will) lead to a decline in GPU rental prices as the downtime allows rivals to reinvigorate their infrastructure.

Without its huge GPU rental income, SpaceX will likely drop back into a sea of red ink, as its AI losses and cash burn exceed that of Starlink’s huge profits. Figure 3 below shows all the rival valuations and returns on a 2028 consensus-basis, and why it makes absolutely no sense for SpaceX to trade at its current valuations.

So yea it looks like OpenAI and Anthropic and the whole AI bubble is in deep deep trouble, know they won’t beat China and have no path to trillion dollar valuations and coincidentally a huge wave of AI doomer hype breaks over the MSM which reinforces their calls for regulation, protection from Chinese AI and possibly a nice taxpayer bailout.

Also watch Corey Doctorow call family blogging on all this doomer nonsense:

VIDEO

Key quotes:

“What an amazing thing, to have worked at the God factory and decided that they had the wrong god. It’s very self-aggrandizing. I’m not saying they do it consciously. I’m just saying it’s very seductive.”30:20

“I think they have AI psychosis. They’ve spent so long going into the bathroom, shutting the door, holding a flashlight under their chin and looking in the mirror going ‘AI,’ scaring themselves silly, that it’s just stuck.”25:05

“If I am inventing God, or stopping people from inventing God, then I am very important.” 28:48

“If you meet someone who’s built a piece of autonomous malicious software in a sandbox to see how well it would work, and it escaped and damaged a server, your first question would be: why are you so bad at making sandboxes? It would not be: why are you so good at making hacking tools?”

“Things can be bad and screwed up and worrisome, and things that we should do something about, without being apocalyptic and without it meaning that there has been a foundational breakthrough in computer science that endangers the whole human race.” 41:29

Big fun. Stay safe, y’all.

Follow and support my work at NatWilsonTurner.com

Related Links

Print Friendly, PDF & Email