There is a shadow hanging over this Fable thing
There is a shadow hanging over this Fable thing
这件事笼罩在 Fable 的阴影之下
Well. I wasn’t quite planning to write this evening, since it’s Friday and Fridays are when I like to code, and when I say code I mean ‘let the agent code while I watch soccer with my friends.’ Recently I’ve been making some fun html games. I actually have another draft post in the barrel about how I think we should see a resurgence of the ‘flash game’ renaissance because it has become so much easier to make fun little games with AI tooling. But in the middle of me thinking about how to make my shitty backrooms-themed shooter play a bit better, the agent went ‘Sorry! This model doesn’t exist any more!’ What the fuck?
好吧。我本来没打算今晚写东西，因为今天是周五，周五是我喜欢写代码的时候——当我说是“写代码”时，我的意思是“让 AI 智能体去写，而我则和朋友们一起看足球赛”。最近我一直在做一些有趣的 HTML 游戏。我其实还有一篇草稿没发，内容是关于我认为我们应该迎来“Flash 游戏”复兴的，因为现在用 AI 工具制作有趣的小游戏变得容易多了。但就在我思考如何改进我那个粗制滥造的“后室”主题射击游戏时，智能体突然弹出一句：“抱歉！该模型已不存在！”这到底是什么鬼？
My first thought was that I needed to re-login. I run a ton of agents in parallel most of the time, so my instinct was that this was just a really really weird limit error. I vaguely knew that Anthropic was thinking about pulling Fable off the subscription plans so I switched to the API. Still nothing.
我的第一反应是需要重新登录。我平时经常并行运行大量的智能体，所以我的直觉告诉我这只是一个非常奇怪的限额错误。我隐约知道 Anthropic 正在考虑将 Fable 从订阅计划中移除，所以我切换到了 API。结果还是不行。
My team built a custom rust agent client, it’s pretty great. But my next thought was ‘o shit the harness bricked’, and I started poking around in Rust, which is a language I barely know even though I’ve ostensibly written tens of thousands of lines of code of it. At which point my friend went ‘the government banned fable.’ What the fuck? But it’s true.
我的团队构建了一个自定义的 Rust 智能体客户端，它非常棒。但我的下一个念头是“糟了，框架崩溃了”，于是我开始研究 Rust 代码——尽管我写过数万行 Rust 代码，但我对这门语言其实知之甚少。就在这时，我的朋友说：“政府封禁了 Fable。”什么鬼？但这是真的。
The US Gov directed Anthropic to disable access to Fable and Mythos to any foreign national anywhere in the world, including those in the US, including Anthropic employees. This is an impossible ask, and the Government knows it, so Anthropic has disabled all access to Fable/Mythos.
美国政府指示 Anthropic 禁止全球任何地方的外国公民访问 Fable 和 Mythos，包括身在美国境内的外国公民，甚至包括 Anthropic 内部的外国籍员工。这是一个不可能完成的任务，政府对此心知肚明，因此 Anthropic 只能彻底关闭了对 Fable/Mythos 的所有访问权限。
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected.
美国政府援引国家安全授权，发布了一项出口管制指令，要求暂停任何外国公民（无论身处美国境内还是境外，包括 Anthropic 的外籍员工）对 Fable 5 和 Mythos 5 的所有访问。该命令的直接后果是，我们必须立即为所有客户禁用 Fable 5 和 Mythos 5 以确保合规。其他所有 Anthropic 模型的访问权限将不受影响。
We received the directive from the government today at 5:21pm (ET). The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or “jailbreaking” Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.
我们于今天下午 5:21（东部时间）收到了政府的指令。信中并未提供有关国家安全担忧的具体细节。据我们了解，政府认为他们发现了一种绕过或“越狱”Fable 5 的方法。我们审查了该技术的演示，发现它被用于识别少量此前已知的轻微漏洞。这些漏洞看起来都相对简单，而且我们发现其他公开可用的模型也能在无需绕过的情况下发现它们。
We have reviewed a report that we believe is the basis of the government’s directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe. We will share more details over the next 24 hours. We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users. However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.
我们审查了一份报告，认为这就是政府指令的依据，并验证了其中展示的能力水平在其他模型（包括 OpenAI 的 GPT-5.5）中广泛存在，并且每天都被负责系统安全的防御者所使用。我们将在未来 24 小时内分享更多细节。我们正在遵守政府的法律指令，并取消所有用户对 Fable 5 和 Mythos 5 的访问权限。然而，我们不同意仅仅因为发现了一个局部的潜在越狱漏洞，就应该成为召回一个已部署给数亿人的商业模型的理由。
A few thoughts about this. Up front, I’m extremely conflicted. I am an AI doomer most days. Having trained many many deep neural networks in my time, I have a deep appreciation for the ways in which optimizers can go wrong. We optimize what we can measure, not what we actually want to achieve. We hope and pray that these are the same thing, but they often aren’t.
关于这件事，我有几点想法。首先，我感到非常矛盾。我大部分时间都是个“AI 末日论者”。在我的职业生涯中，我训练过许多深度神经网络，我深刻理解优化器可能出错的方式。我们优化的是我们能衡量的东西，而不是我们真正想要实现的目标。我们祈祷这两者是一回事，但事实往往并非如此。
Corporations are ALSO optimizers, so of course they are optimizing for ‘get money as fast as possible’ — the thing we can measure — despite many of the people building it being like ‘hey yea this is really dangerous,’ which is of course a fantastic parable for the whole AI alignment debate.
企业也是优化器，所以它们当然是在优化“尽可能快地赚钱”——这是我们能衡量的东西——尽管许多构建它的人会说“嘿，这真的很危险”，这当然是整个 AI 对齐辩论的一个绝佳寓言。
But also, there is a massive shadow hanging over this whole thing. If any other government in the history of the United States took this step, there would be good reason to at least give that government the benefit of the doubt. But this government has shown itself to be petty and corrupt in ways that continue to completely and totally astound me in its openness and creativity. Is this coming from an actual desire to regulate AI? A better question: does anyone in this government who knows anything about AI actually have the ear of people who make these decisions? I would bet against it!
但同时，这件事笼罩着巨大的阴影。如果美国历史上的任何其他政府采取这一举措，我们至少有充分的理由给予政府一定的信任。但本届政府表现出的卑劣和腐败，其公开程度和创造力简直让我彻底震惊。这是出于真正监管 AI 的愿望吗？更好的问题是：政府里那些真正懂 AI 的人，真的能影响那些做决策的人吗？我敢打赌，不可能！