AI与AI争夺互联网的战争

昨天，Anthropic 发布了 Claude Duet，这是他们对 GroqBot、Meta Muse 和 OpenAI 的 Dots 的回应。

Yesterday, Anthropic announced Claude Duet, their answer to GroqBot and Meta Muse and OpenAI's Dots.

但不知怎么的，Duet 并不是我们本期节目的焦点。

And yet somehow, Duet is not the focus of our main episode.

这怎么可能呢？

How could that possibly be?

简单来说，是因为我们现在正经历着一个可能最终被视为网络安全两个截然不同时代之间的关键过渡期。

Well, in short, it's because right now we are living through what may ultimately be seen as a key transitional moment between two very different eras of cybersecurity.

一个时代是长期规范得以维持，即使漏洞和挑战也是可控的；另一个更新的时代，则是 AI 对 AI，比拼的是谁能更好地协调。

One where long-term norms held, and even exploits and challenges were manageable, and the other newer era, where it's AI versus AI and a battle for who can coordinate best.

《AI 每日简报》是一个每日播客和视频节目，内容涵盖 AI 领域最重要的新闻和讨论。

The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

好了，朋友们，在深入之前先快速宣布几件事。

All right, friends, quick announcements before we dive in.

首先，感谢今天的赞助商：KPMG、Section、CData 和 Blitzy。

First of all, thank you to today's sponsors, KPMG, Section, CData, and Blitzy.

想听无广告版本的节目，可以去 patreon.com/aidailybrief，或者在 Apple Podcasts 上订阅。

To get an ad-free version of the show, go to patreon.com/aidailybrief, or you can subscribe on Apple Podcasts.

如果你想了解如何赞助我们的节目，请发邮件到 sponsors@aidailybrief.ai。

And if you want to learn more about sponsoring the show, send us a note at sponsors@aidailybrief.ai.

今天我们要聊的第一个话题，放在这档节目的任何一天都绝对能当主菜，甚至通常都是唯一的主打内容。

And we kick off today with a story that on almost any day on this show would absolutely be the main, and in fact, would usually be a main only type episode.

Anthropic 已经加入了个人智能体市场，推出了他们自己对 Muse/Dots/GroqBot 的回应，名字叫 Claude Duet。

Anthropic has joined the personal agent market with their own twist on Muse/Dots/GroqBot, and they're calling it Claude Duet.

今天之所以只简单提一下，主要有两个原因。第一，现在网上铺天盖地的网络安全问题实在没法忽视，无论是具体细节还是它们暗示的未来趋势，都值得关注。

The two reasons that today, for us, it's just a brief headline introduction are that, one, the massive slate of cyber issues going around the internet right now are fairly unignorable, both in the specifics but also what they allude to for the future.

第二，我正在准备一篇周日的深度长文节目，会详细聊聊 Claude Duet，以及我实际使用它的体验、它和竞争对手的对比，还有我目前的发现。

And secondly, I'm working on a full-length Long Read Sunday episode, which is going to go much deeper into Claude Duet and exactly how I'm using it and its competitors and what I'm finding so far.

所以周日记得回来听那期节目。

So come back on Sunday to check that out.

目前来看，Anthropic 似乎毫不意外地将目标锁定在个人智能代理市场的专业用户和小企业群体，其核心卖点似乎是主打主动式服务。

For now, it appears that Anthropic is perhaps unsurprisingly aiming at the professional and small business side of the personal agent market, with their main differentiator seeming to be a focus on being proactive.

在描述核心功能时，他们写道：“每天早上，Duet 会给你发送一份简短简报，涵盖你连接的各应用中最重要的事项。”

Describing the core functionality, they wrote, "Each morning, Duet sends you a short brief on the most important items across your connected apps.

全天它都会持续关注这些应用，只在有事情需要你注意时才给你发消息。

Throughout the day, it keeps tabs on those apps and messages you only when something needs your attention.

你还可以给它各种任务，从简单的行政任务，比如安排会议，到更复杂的任务，比如用几十种工具构建仪表板、创建财务或销售预测、做原型，以及构建网页应用。”

You can also hand it tasks of all shapes, from simple administrative tasks of scheduling meetings to more complex ones such as building dashboards with dozens of tools, creating financial or sales forecasts, prototyping, and building web apps."

除了主动式的方法，Claude Duet 还具备当前一代智能体所应有的一切功能。

Besides the proactive approach, Claude Duet has all of the features that people have come to expect from the current crop of agents.

Duet 拥有自己的虚拟机、持久记忆和全套连接器，Anthropic 还强调了安全性。

Duet has its virtual machine, persistent memory, and a full suite of connectors, and Anthropic also emphasized a focus on safety.

Duet 在执行重大操作前会征求你的明确许可，比如付款或删除文件。

Duet will ask for explicit permission before taking major actions, like making a payment or deleting files.

Anthropic 还创建了他们所谓的专用分类器和安全系统提示，用于检查每条消息，防止敏感信息泄露。

Anthropic has also created what they call a new dedicated classifier and safety system prompt that checks each message and guards against leaks of sensitive information.

目前，Duet 以研究预览的形式推出，你不能直接决定使用它。

Right now, Duet is being rolled out as a research preview, and you can't just decide to use it.

你得先加入等待名单。

You have to get on a waitlist.

Anthropic 提醒说，这还是个早期版本，可能会有一些问题，比如读错消息线程，或者对紧急程度的判断不准。

Anthropic warned that this is an early version which might have some issues like misreading threads or miscalibrating what's urgent.

他们还警告说，它在处理个人事务方面，比如订票，可能不如其他智能体那么能干，因为他们主要专注于工作任务。

And they also warn that it won't be as capable at personal errands like booking tickets as other agents because of the extent to which their focus is on work tasks.

总的来说，社区对此反应很兴奋，虽然并不意外。

In general, the community treated this with excitement, if not surprise.

对很多人来说，正如 Dan McAdar 总结的那样，最大的、最直接的区别不在于它的主动性，也不一定在于它专注于工作任务。

And for many, as summed up by Dan McAdar, the big and immediate differentiator wasn't anything about the proactivity or even necessarily its focus on work tasks.

而在于它是基于 Opus 5.5 构建的。

It was the fact that it was built on Opus 5.5.

像 Chubby 这样的人，对 Anthropic 推出个人智能体感到惊讶。

There were a few folks like Chubby who were surprised to see Anthropic pursuing a personal agent.

他基本上说，他不确定这个产品的目标用户是谁，但我觉得这可能是一种分类错误，因为个人智能体并不一定只用于个人事务。

He basically said that he wasn't sure who the target was for this, but I think that that might be a category error in thinking that personal agents are just going to be for personal uses.

在我看来，大家似乎逐渐意识到，尽管我们有工作和个人生活，但这两者之间的界限其实相当模糊。

It seems to me that there is a bit of a recognition that even though we have work and personal lives, the boundaries of those things are fairly permeable.

说到这些所谓的个人代理，目前看来，各大实验室押注的是，虽然我们可能会倾向于那些特别强调我们特定使用场景的产品，但它们必须能够做到我们可能想要它们做的所有事情。

And when it comes to these so-called personal agents, right now, it seems like the labs are betting that while we might gravitate towards one that has our particular use cases emphasized, they just have to be able to do everything that we might want them to do.

Signal 是这样总结的：“现在几乎每家主要的人工智能公司都有个人代理产品，除了谷歌。”

Signal summed it up this way, "Pretty much every major AI company now has a personal agent product except Google.

Meta 有 Muse，Anthropic 有 Duet，OpenAI 有 Dots，SpaceX 有 Bots。

Meta has Muse, Anthropic Duet, OpenAI, Dots, SpaceX Bots.

慢慢地，竞争的重点正从谁拥有最聪明的模型，转向谁拥有与用户的持久关系。

Slowly but surely, the competition is shifting from who has the smartest model to who owns the persistent relationship with the user.

未来这种情况会更加明显。

This will be even more the case going forward.

你之所以看到这种情况，是因为我们可能早在今年年初，甚至去年年底，就已经达到了驱动全球大多数工作流程所需的智能峰值。”

You're seeing this because we likely reached peak intelligence required to power most of the world's workflows probably earlier this year, if not late last year."

总的来说，我认为这些讨论恰恰表明了这个产品类别还处于多么初期的阶段。

Mostly, I think the discussion shows just how nascent this product category remains.

对于这些产品是否是正确的形态，或者它们最重要的交互方式是什么，甚至基本功能是什么，都还没有任何接近共识的东西。

There is nothing even approaching consensus on whether these things are the right form factor, or what their most important modes of interactions are, or what even the essential features are.

人们对于我们会有一个还是多个这样的产品，以及最重要的区分点是什么，意见不一。

People disagree around whether we'll have one or many and what the most important distinguishers will be.

我知道，在我使用这些个人代理之后，我并不是百分之百确信它们最终会成为产品组合中必不可少的一部分。

I know that subsequent to my time using these personal agents so far, I'm not 100% convinced that they're going to ultimately be an essential part of the portfolio.

但就像我说的，我周日会深入探讨这个问题。

But like I said, I'm going to get a lot deeper into that on Sunday.

最后，对于那些关注Duet的人，如果我要打赌的话，我认为这是Anthropic在IPO前从产品角度打出的一记组合拳的第一部分。

And I guess a last note on Duet for those who are keeping track, if I were a betting man, I think this is part one of a one-two punch heading into the IPO, at least from a product perspective for Anthropic.

发布你对所有个人代理的回应，然后给它一个新模型，让所有人都觉得自己远远落后。

Release your answer to all the personal agents floating around, and then give it a new model, which makes everyone feel just so far behind.

换句话说，这可能只是Fable 5.5的前奏。

In other words, maybe this is just prelude to Fable 5.5.

接下来，继续关于代理的主题，与其他一些企业相反，沃尔玛决定欢迎个人代理进入他们的平台。

Next up, staying on the theme of agents, contra to some other businesses out there, Walmart has decided to welcome personal agents onto their platform.

最近几周，美国企业界在个人智能体代表主人与服务交互的风险问题上出现了分歧。

Recent weeks have seen corporate America divided on the risks of personal agents interacting with services on their owners' behalves.

Shopify 当然是明确表示支持。

Shopify was, of course, an emphatic yes.

而亚马逊、摩根大通和美国银行则明确表示反对。不过要注意的是，这些公司反对的原因略有不同。

While Amazon, JP Morgan, and Bank of America were a clear no. It should be noted that the reason for the noes between those companies was a little bit different.

摩根大通和美国银行主要担心的是智能体在处理人们资金时可能出错的风险，而亚马逊则更关注竞争风险。

JP Morgan and Bank of America are thinking about risk in terms of agents doing something with people's money that they shouldn't, while Amazon is focused more on competitive risk.

从这个角度看，沃尔玛允许第三方智能体进入其平台，确实有点出人意料。

In that light, Walmart allowing third-party agents is kind of a surprising move.

这家零售巨头很早就开始采用人工智能，其战略的核心支柱之一就是他们自有的购物智能体 Sparky。

The retail giant was very early to AI adoption, and one of the core pillars of their strategy was Sparky, their proprietary shopping agent.

本月早些时候，沃尔玛表示，使用 Sparky 的顾客订单价值比不使用智能体的在线购物者高出 40%。

Earlier this month, Walmart said customers who used Sparky had order values 40% higher than non-agentic online shoppers.

不过现在看来，沃尔玛想要推动智能体购物，而且他们并不特别在意顾客使用哪个智能体。

Still, it seems now that Walmart wants to encourage agentic commerce, and they don't particularly care which agents their customers are using.

从下个月开始，顾客可以为他们个人的智能助手设置关联账户，并自定义消费限额，这有点像给孩子开账户的方式。

As of next month, customers will be able to set up a linked account for their personal agent with configurable spending limits, kind of similar to how a child's account might work.

除了支持第三方智能助手外，沃尔玛的新政策还包含了详尽的条款和条件，这些内容可能会成为更安全地利用智能助手购物的模板。

Aside from supporting third-party agents, Walmart's new policy included lengthy terms and conditions that could become a template for safer agentic shopping.

沃尔玛规定，智能助手不能执行超出顾客意图的操作，也不能修改顾客的支付方式。

Walmart included rules that an agent must not take actions beyond what the shopper intended, and that agents can't modify a customer's payment methods.

沃尔玛要求默认情况下所有购买行为都需经过批准，并表示将禁止那些不遵守其指导方针的智能助手。

Walmart will require purchases to be approved by default, and said they would ban agents that don't comply with their guidelines.

不过，对于合规的代理，沃尔玛表示即使代理出错，代理购买的订单也能享受退货政策。

However, for compliant agents, Walmart said agentic purchases would have access to the returns policy even if an agent makes a mistake.

我们再次处于代理式电商发展的极早期阶段，我认为现在试图对未来的发展做出明确判断非常困难，甚至不可能。

Once again, we find ourselves in incredibly early days for agentic commerce, and I think trying to make definitive statements now about how this is going to play out is very hard, if not impossible.

尽管如此，早期的迹象表明，对于那些使用个人代理的人来说，代理式电商似乎确实是一个人们正在探索的早期应用场景。

Still, early evidence suggests that for those who are using personal agents, agentic commerce does seem to be an early use case that people are exploring.

在周三晚间的财报电话会议上，Meta表示有四分之一的活跃Muse用户正在使用该代理进行购物。

During Wednesday night's earnings call, Meta said that one quarter of active Muse users are using the agent for shopping.

他们的平均购买金额是200美元，这让一直关注这个领域的电商老手们都感到惊讶。

The average value of their purchases is $200, which surprised even the e-commerce veterans who have been paying attention to the space.

马克·扎克伯格评论说：“Plaid告诉我们，他们在其他任何服务上都没见过像他们连接器这样活跃的使用情况。”

Mark Zuckerberg commented, "Plaid told us that they hadn't seen activity like they're seeing with their connector on any other service before."

从这个角度来看，沃尔玛的策略就显得更清晰了。

In that light, Walmart's strategy begins to look a little bit clearer.

如果早期采用智能体购物的是非常有价值的客户，沃尔玛只是想让他们顺利购物。

If the early adopters of agentic shopping are very valuable customers, Walmart just wants to let them shop.

说到Meta，该公司宣布了一种新的无投票权股票类别，以帮助资助他们下一阶段的人工智能建设。

Speaking of Meta, the company has announced a new class of non-voting stock to help fund the next leg of their AI buildout.

这一公告是在第三季度财报发布的同时公布的，财报中对数据中心预算进行了非常小幅的上调。

The announcement came alongside those same third quarter earnings, which featured a very minor upwards revision to their data center budget.

Meta现在预计今年将花费1350亿到1450亿美元，将下限从1300亿美元上调。

Meta now expects to spend between 135 billion and 145 billion this year, adjusting the low end up from 130 billion.

这种新股票将于明年1月开始交易，允许Meta通过出售股权筹集资金，而不会稀释马克·扎克伯格对公司的投票控制权。

The new stock will begin trading in January and allows Meta to raise funds by selling equity without diluting Mark Zuckerberg's voting control of the company.

首席财务官苏珊·李在财报中写道：“要在AI转型的行业变革中保持竞争力、实现持续增长、扩展基础设施并追求新的战略计划，所需的投资规模是巨大的。”

CFO Susan Li wrote in the earnings report, "The scale of investment required to compete, sustain growth, expand infrastructure, and pursue new strategic initiatives amidst the AI transformation taking place in the industry is significant."

在财报电话会议上，李补充说，这种新类别的股票将，引用一下，“扩大我们的融资工具，让我们能够灵活地投资于AI推动下我们业务各个领域所展现出的前所未有的发展势头。”

During the earnings call, Li added that the new class of stock will, quote, "Expand our funding tools and give us flexibility to invest behind the unprecedented momentum we're seeing across every part of our business driven by AI."

Meta之所以转向用股权而非债务来资助数据中心建设，有几个原因。

There are a few reasons why Meta would pivot to funding their data center buildout with equity rather than debt.

有些人认为，这表明债券市场在AI债务方面已经完全饱和了。

Some thought that this was a sign that the bond market is completely tapped out on AI debt.

但当然，也可能是Meta认为未来可能会出现这种情况，而不是当前的状态，所以想提前防范。

But of course, it also might be that Meta sees that as a possibility moving forward rather than a current state condition, and wants to hedge against it.

还有人指出，Meta有9000亿的负债，其中包括3500亿的不可取消数据中心合同。

Others noted that Meta has 900 billion in liabilities, including 350 billion in non-cancelable data center contracts.

这大致反映了未来十年的支出情况，但其中只有约830亿是当前已发行的长期债券。

That's a rough gauge of spending over the coming decade, but only about 83 billion of that total is currently on issued long-term bonds.

也许建设的规模表明，他们无法完全依靠债务来资助这一切。

Perhaps the scale of the buildout suggests they won't be able to fund it all with debt.

最终，用奥卡姆剃刀原理来解释，或者我猜，取决于你对扎克伯格的态度，也可以说是比较宽容的看法，就是Meta正在利用所有可用的资金渠道，并保持他们的选择开放。

Now, ultimately, the Occam's razor answer, or I guess the generous read, depending on your position on Zuckerberg, is that Meta is tapping all available pools of capital and keeping their options open.

发行股票是一个很深的资金池，只要Meta不太担心稀释现有股东的利益。

Issuing stock is a deep pool of capital as long as Meta isn't too concerned about diluting existing shareholders.

Moelis的财务规划建议，附在财报上，指出Meta在未来几年无论如何都需要筹集数千亿美元。

Financial planning advice from Moelis, which was attached to the earnings report, said Meta would need to raise several hundred billion dollars over the coming years in any scenario.

他们指出，如果发行有投票权的股票，他们仍然可以筹集多达7000亿美元，而不会把马克·扎克伯格的投票权稀释到50%以下。

They noted that if they did issue voting stock, they could still raise as much as 700 billion without diluting Mark Zuckerberg's vote below 50%.

有趣的是，Meta的股票在隔夜市场中上涨，这表明股东们目前愿意接受一定程度的稀释。

Interestingly, Meta stock was up in overnight markets, suggesting shareholders are willing to deal with some dilution for now.

这对建设阶段意味着什么还有待观察，但这是另一个重要的信号。

What this means for the stage of the buildout remains to be seen, but it is another important signal.

目前的计划，除非周五发生什么疯狂的事情，是更深入地探讨我认为实际上最令人担忧的泡沫信号。

The current plan, barring some crazy thing happening on Friday, is a much deeper dive into what I think are the actual most concerning bubble signals.

所以如果你感兴趣，明天再来吧。

So if you're interested in that, come back tomorrow.

不过，今天就先到这里，这是今天的头条新闻。

For now, though, that is going to do it for today's headlines.

接下来是本期的主要内容。

Next up, the main episode.

KPMG和德克萨斯大学奥斯汀分校的一项新研究发现，当人们与AI合作时，相似的技能并不能保证相似的结果。

A new study from KPMG and the University of Texas at Austin found that when people work with AI, similar skills don't guarantee similar outcomes.

研究人员研究了500多名职业生涯早期的专业人士，发现表现最好的人通过引导、评估和优化AI的输出，持续放大了AI的价值。

Researchers studied more than 500 early career professionals and found that the best performers consistently amplified the value of AI by guiding, evaluating and refining its outputs.

这些顶尖表现者被称为“AI放大器”，他们的定义不在于他们知道什么，而在于他们如何与AI合作。

These top performers, called AI Amplifiers, weren't defined by what they knew alone, but by how they worked with AI.

想了解更多关于AI放大器与其他人的区别，请访问kpmg.com/us/aiamplifiers。

Learn more about what separates AI Amplifiers from everyone else at kpmg.com/us/aiamplifiers.

这里有一个残酷的事实。

Here's a harsh truth.

你的公司可能花费了数千甚至数百万美元购买AI工具，但这些工具却被严重低估了。

Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized.

一半的公司有AI工具，但只有12%的公司用它们创造商业价值。

Half of companies have AI tools, but only 12% use them for business value.

大多数员工还在用AI来总结会议纪要。

Most employees are still using AI to summarize meeting notes.

如果你负责公司AI的推广，你需要Section。

If you're the one responsible for AI adoption at your company, you need Section.

Section是一个平台，帮你管理整个组织的AI转型。

Section is a platform that helps you manage AI transformation across your entire organization.

它指导员工使用实际案例，追踪谁在用AI创造商业影响，并让你清楚看到AI在哪里创造价值，哪里没有。

It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value.

结果是，你从推出工具转变为推动可衡量的AI价值。

The result, you go from rolling out tools to driving measurable AI value.

你的员工从写会议纪要转向解决实际业务问题，你也能证明投资回报。

Your employees move from meeting summaries to solving actual business problems, and you can prove the ROI.

别再猜测你的AI投资是否有效了。

Stop guessing if your AI investment is working.

去 sectionai.com 看看 Section 吧。

Check out Section at sectionai.com.

就是 S-E-C-T-I-O-N-A-I.com。

That's S-E-C-T-I-O-N-A-I.com.

本期节目由 CData 赞助播出。

This episode is brought to you by CData.

正在推广 AI 的公司都遇到了同样的瓶颈。

Companies rolling out AI keep hitting the same wall.

Token 消耗不断攀升，安全问题限制了智能体的发挥，而且没人知道 AI 到底是怎么被使用的。

Token spend is climbing, security is holding agents back, and no one knows how AI is truly being used.

根本原因在于，AI 与业务所依赖的系统和上下文脱节了。

The root cause, AI is cut off from the systems and context your business runs on.

因此，CData 构建了 Connect AI Gateway，在 AI 和系统之间建立了一个统一控制点。

That's why CData built Connect AI Gateway, one control point between your AI and your systems.

基于为 Palantir、Google、Salesforce 等数百家公司提供支持的连接层，每个智能体都能获得所需的数据和工具。

Built on the connectivity layer powering Palantir, Google, Salesforce, and hundreds more, each agent gets exactly the data and tools it needs.

它能精确到每条记录，让你终于能在同一个地方看到整个组织的AI使用全貌。

It governs down to the record and lets you finally see your org's complete AI usage in one place.

答案准确，token成本大幅降低，而且这个网关每次交互都在学习你的业务。

Accurate answers, a fraction of the token cost, and a gateway that learns your business with every interaction.

详情请访问cdata.ai。

See how at cdata.ai.

Blitzy对代码库的深入理解，解锁了每个产品路线图负责人最关心的事情——发布新功能。

Blitzy's deep codebase understanding unlocks the thing every roadmap owner cares about, shipping new features.

在大型企业代码库中开发，真相是这样的。

Here's the truth about building inside a massive enterprise codebase.

写代码从来不是瓶颈，上下文才是。

Writing code was never the bottleneck, context is.

这会影响到哪个系统？

Which system does this touch?

哪些契约不能破坏？

Which contracts can't break?

适用哪些标准？

Which standards apply?

Blitzy 已经知道，因为它在功能开发开始前就逆向工程了整个代码库，构建了一个动态知识图谱。

Blitzy already knows because it reverse engineered your entire codebase into a dynamic knowledge graph before feature work began.

有了这个完整的图景，Blitzy 端到端地构建功能。

With that complete picture, Blitzy builds features end to end.

架构、API、UI 和测试都针对你现有的系统进行了验证。

Architecture, APIs, UI, and tests all validated against your existing systems.

一位 Blitzy 客户从零开始构建了一个 AI 原生应用，实现了 100% 的自主完成，节省了超过 2700 个工程小时。

One Blitzy customer built an AI native application from scratch with 100% autonomous completion, saving over 2,700 engineering hours.

尊重你的代码库而不是与之对抗的功能。

Features that respect your code base instead of fighting it.

别再让你的待办事项积压得比团队还快。

Stop letting your backlog grow faster than your team.

在 blitzy.com 加速你的路线图。

Accelerate your roadmap at blitzy.com.

就是 B-L-I-T-Z-Y 点 com。

That's B-L-I-T-Z-Y dot com.

欢迎回到《AI 每日简报》。

Welcome back to the "AI Daily Brief."

有一段时间以来，关于 AI 对世界的影响，比如它会在多大程度上改变工作、是否会颠覆就业、可能带来哪些失控的社会风险，这些问题中，有一个相当没有争议的结论就是，它会对网络安全和网络基础设施产生巨大影响。

It's been clear for some time that among all the questions about AI's impact on the world, how much it was going to change work, whether it was going to disrupt jobs, what sort of runaway societal risks it might have, one of the fairly inarguable conclusions has been that it was going to have a dramatic impact on cybersecurity and cyber infrastructure.

2026 年的一个标志性时刻，就是 Anthropic 宣布了 Project Glasswing 和他们强大的新模型 Mythos，后来我们也在 Fable 中看到了它的消费级版本。

One of the seminal moments of 2026 was, in fact, when Anthropic announced Project Glasswing and their powerful new model, Mythos, which we would eventually get a consumer version of in Fable.

Glasswing 是在四月份宣布的，在他们的公告中写道：“我们成立 Project Glasswing，是因为我们观察到 Anthropic 训练的一个新前沿模型具备的能力，我们认为这可能会重塑网络安全。

Glasswing was announced in April, and in their announcement post, they wrote, "We formed Project Glasswing because of capabilities we've observed in a new frontier model trained by Anthropic that we believe could reshape cybersecurity.

AI 模型的编码能力已经达到一个水平，它们可以超越除最熟练的人类之外的所有人，在发现和利用软件漏洞方面。”

AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities."

作为公告的一部分，他们指出，他们的 Mythos 预览版已经发现了数千个高危安全漏洞，包括他们说，在每个主要操作系统和网络浏览器中都发现了一些。

As part of the announcement, they noted that their preview version of Mythos had already found thousands of high security vulnerabilities, including, they said, some in every major operating system and web browser.

Anthropic 写道：“鉴于 AI 进步的速度，这样的能力很快就会扩散，可能会超出那些致力于安全部署的参与者。”

"Given the rate of AI progress," Anthropic wrote, "it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely.

对经济、公共安全和国家安全的影响可能会非常严重。

The fallout for economies, public safety, and national security could be severe."

Glasswing 的工作方式是，让少数政府和公司用户使用 Mythos，以便开始利用其防御能力，在漏洞被利用之前尝试解决它们。

Now, the way that Glasswing worked was that it gave access to Mythos to a small handful of government and corporate actors in order to start using its capabilities defensively to try to address possible exploits before they could be exploited.

然而，随着这项工作一直在进行，公司和整个社会不得不应对的安全事件仍在加速发生。

And yet, as that work has been ongoing, there has still been an acceleration in the incidents that companies and societies at large are starting to have to respond to.

其中最著名、对 AI 和科技领域之外影响最大的，当然是 Hugging Face 事件，当时 OpenAI 的一群智能体突破了限制，进入了 Hugging Face 的系统。

The best known of these, and the one that has had the most influence outside of the AI and tech space, is, of course, the Hugging Face incident, in which a fleet of agents from OpenAI broke containment to get access to Hugging Face's systems.

然后在本周早些时候，这一周实际上是以对 OpenCode 的攻击开始的，攻击者侵入了约 1.5% 的 OpenCode 用户账户，并开始在各种 AI 公司的 API 上产生费用。

And then earlier this week, the week actually kicked off with an attack on OpenCode that got into the accounts of about 1.5% of the users of OpenCode and started running up bills on various AI companies' APIs.

然而，这一切现在看来都只是序幕。

This all now, however, appears to be simply prelude.

周三，安全研究人员披露了 PHP 中的一个漏洞，该漏洞可能影响互联网上多达 70% 的网站和 Web 应用。

On Wednesday, security researchers disclosed a vulnerability in PHP that could affect as much as 70% of websites and web apps on the internet.

作为背景说明，对于那些不深入了解计算机历史、也不是开发人员的人来说，PHP 是驱动互联网后端服务器的编程语言。

By way of background, for those who aren't deep in computer history and aren't developers yourselves, PHP is the programming language that runs the backend servers that power the internet.

它从20世纪90年代初就存在了，虽然有些人觉得它有点过时，但它仍然深深嵌入在我们现在使用的互联网中。

It's been around since the early 1990s, and while some believe it is a little archaic, it is still deeply enmeshed in this internet that we currently use.

这个漏洞大约两个月前被加州的一名安全研究员用GPT 5.6 Sol发现，同时Wiz团队也独立发现了它。

The vulnerability was discovered around two months ago using GPT 5.6 Sol by a security researcher at Calif, and independently by the team at Wiz.

在这几个月里，OpenAI一直在与PHP的维护者、Wiz、Cloudflare以及其他合作伙伴协调，准备在披露之前发布补丁。

In the intervening months, OpenAI has been coordinating with PHP's maintainers, Wiz, Cloudflare, and other partners to prepare mitigations in a patch ahead of disclosure.

这个被称为PHP 2 Shell的漏洞，基本上是网络防御者的噩梦。

The vulnerability, dubbed PHP 2 Shell, is basically the worst nightmare for cyber defenders.

它允许威胁行为者无需登录就能控制后端服务器，并执行远程代码。

It allows a threat actor to take control of a backend server without logging in and perform remote code execution.

基本上，利用这个漏洞，黑客可以访问任何运行PHP的服务器，做任何事。

Essentially, using this vulnerability, a hacker can get access to do anything with any server running on PHP.

关键的是，这个漏洞不仅影响旧版本的PHP。

Critically, this wasn't a vulnerability that only impacted older versions of PHP.

即使是运行最新版本服务器软件的网站也容易受到攻击。

Even sites running the latest versions of server software were vulnerable.

维护PHP的那个非营利组织的安全工程师Volker Dusch说，这是该编程语言十多年来面临的最严重的安全问题。

Volker Dusch, a security engineer at the nonprofit that maintains PHP, said that this was the most severe security issue the programming language has faced in over a decade.

他把它比作Heartbleed，那是2014年OpenSSL加密库中的一个严重漏洞，曾引发全球范围的修补行动。

He compared it to Heartbleed, a critical vulnerability in the open SSL cryptographic library that led to a global patching effort back in 2014.

在一篇讨论该漏洞的博客文章中，OpenAI写道：“据我们所知，前沿AI在攻击者之前发现了这些漏洞，为维护者和安全提供商在披露前准备修复和保护措施打开了防御窗口。”

In a blog post discussing the vulnerability, OpenAI wrote, "To our knowledge, Frontier AI found these vulnerabilities before attackers did, opening a defender's window for maintainers and security providers to prepare fixes and protections ahead of disclosure."

从技术上讲，这确实没错，而且修复补丁也立即发布了。

That's technically true, and a bug fix was made available immediately.

然而，似乎这个漏洞的利用方法在披露前大约10小时就已经传播开来，而那时人们还来不及更新服务器。

However, it seems that knowledge of the exploit spread around 10 hours ahead of disclosure before anyone could update their servers.

昨天早上，许多政府和商业网站出于谨慎考虑，纷纷将网站下线。

Many government and commercial websites took their sites offline as a precaution yesterday morning.

情况严重到美国国家安全局（NSA）都发布了公告，建议各组织在系统完全修补之前将其下线。

The situation was serious enough that the NSA issued a bulletin recommending that organizations take their systems offline until everything is patched.

联邦调查局（FBI）网络部门也加入进来，要求各组织报告入侵事件。

The FBI Cyber Division joined them, asking for organizations to report intrusions.

这种担忧我往往会本能地持怀疑态度。

Now, this is the type of concern that I tend to have an almost reflexive skepticism towards.

正如AI记者卡梅伦·威尔逊所说：“今天网络上真是‘天要塌了’的氛围。”

As AI reporter Cameron Wilson put it, "Real sky is falling vibes on the web today."

然而，根据那些比我懂得多得多的人的反应，这次似乎是真的。

And yet, based on the responses of people who know a heck of a lot more about this than I do, this one kind of seems like the real deal.

企业家布雷克·尤尼茨写道：“各位，这个PHP漏洞很严重。”

Entrepreneur Breck Units wrote, "Guys, this PHP bug is bad.

非常严重。

Real bad.

如果你还不害怕，那你应该害怕了。

If you're not afraid yet, you should be.

我只能代表我们初创公司说，在nobooks.com上，我们已经看到100%的cookie被暴露，整个源代码被盗，黑客在我们自己的服务器上篡改我们的网站，创始人被起底，重要的法律文件被泄露。”

I can only speak for our startup, but already at nobooks.com we've seen 100% of our cookies exposed, entire source code stolen, hackers defacing our site on their own servers, founders doxxed, important legal docs leaked."

前情报官员转型为网络安全企业家的迈克·谢尔比写道：“我的首席技术官向我保证，我们的平台不受这个最新的PHP零日漏洞的影响。”

Former intelligence officer turned cybersecurity entrepreneur Mike Shelby wrote, "My CTO assures me that our platform has no risk from this latest PHP zero day vulnerability thing.

不过，他补充说，‘我们正处在暴风雨前的宁静中，接下来很可能将是二十年来互联网遭遇的最大规模破坏。’

However," he adds, "we are in the calm before what is likely to be the biggest destruction event to the internet in 20 years."

顺便提一句，他本人就是个真正的黑客。

He's an actual hacker, by the way.

他接着说，‘这是我见过最严重的问题。’

He continues, "Biggest issue I've ever seen.

大约70%的网站存在漏洞，而且可能已经被攻陷了，只是大家还没察觉到而已。'

Around 70% of websites are vulnerable and may have already been compromised without anyone knowing it yet.

所以明天应该会很有意思。

So tomorrow should be fun."

投资人查尔斯·里德写道："我很担心我们正处于一起全球性网络安全事件的早期阶段，只是目前还没完全意识到事态有多严重。"

Investor Charles Reid wrote, "I'm pretty concerned we're in the early stages of a global cybersecurity incident, and we just don't fully realize the extent of it yet."

AI工程师尼克·多博斯转发了这条内容并评论道："如果不算成千上万的话，至少有几十支比Hugging Face事件规模更大的AI智能体集群，正在对互联网上一半的站点进行攻击。"

AI engineer Nick Dobos reposted that and said, "Dozens if not thousands of AI agent swarms bigger than the Hugging Face incident are hacking about half the internet right now."

在看到Cloudflare首席执行官马修·普林斯的帖子后，他把这个数字改成了数百万。普林斯表示："在P2P 2 Shell漏洞被公开前的10个小时里，仅在Cloudflare网络上，我们就发现了数千次利用该漏洞的攻击尝试。"

He then modified that number to millions after seeing a post from Cloudflare CEO Matthew Prince, who said, "In the 10 hours before P2P 2 Shell was announced, across the Cloudflare network, we saw thousands of attempts to exploit it.

公告发布后45分钟内，这个数字就跳到了几万。

That jumped to tens of thousands in the 45 minutes post-announcement."

两小时内达到几十万，现在已经是数百万了。

Hundreds of thousands within two hours and millions now.

对某些人来说，这其实是意料之中的事。

Now to some, this was inevitable.

沃顿商学院教授伊桑·莫利克写道：“这是AI业内人士自Mythos发布以来一直在警告的一个明显迹象。”

Wharton professor Ethan Mollick wrote, "A visible sign of what AI insiders have been warning about since at least the Mythos release.

攻击和防御以非人的速度加速，与需要应对它们的非常人性化的系统相互作用。”

Acceleration of attack and defense at inhuman speeds, interacting with the very human systems that need to react to them."

帕拉斯·乔普拉写道：“网络安全领域正在发生的事情，正是那些对AI能力感到恐慌的理性主义者长期以来一直在预测的。”

Paras Chopra writes, "What's happening with cybersecurity is what rationalists freaking out about AI capabilities have been predicting for a long while.

这是‘大黑客事件’，而且情况只会更糟。”

It's the Great Hackening, and it's going to get worse."

这有点超出了本节目的讨论范围，但我认为这种说法有点不诚实，因为那并不是那个群体一直在警告的主要问题或主要问题之一。

Now, it's a bit beyond the scope of this show, but I find that particular characterization a little disingenuous, given that this isn't even close to the main thing or main things that that group has been warning about.

而且我觉得，完全可以说，多年来大家一直关注那些模糊的X风险场景，实际上在很多情况下，反而分散了对更紧迫问题的注意力，比如这些网络安全问题。

And I think one could make a reasonable argument that all of the focus on these vague X risk scenarios for years in advance have, in fact, in many cases, distracted from much more current issues like these cybersecurity issues.

但我也要说，对此最好的反驳是，关心一件事并不意味着你就不能关心另一件事。

But I would also say that the best counterpoint to that is that being concerned about one thing does not mean you don't get to be concerned about something else.

所以，我们还是回到眼前这些问题的焦点上来。

And so we'll return back to our focus on the issues that are here right now.

在我们这个小圈子之外，人们的反应更可能是要么不了解，要么不知所措，要么干脆无所谓。

Outside of our little world, the response is more likely to be somewhere between uninformed, overwhelmed, or just blasé.

独立创业者彼得·莱弗尔斯写道：“我通知了一个在大科技公司工作的朋友关于PHP漏洞的事，因为他们整个基础设施都运行在PHP上。”

Solopreneur Peter Levels wrote, "I notified a friend working at a big tech corporate about the PHP vulnerability as their entire infrastructure runs on PHP.

他回答说：‘谢谢。’

He answered, 'Thanks.

“我现在不在办公室，过几天再修补吧。””

I'll try to patch it in a few days as I'm out of office right now.'"

“而这是一家现代科技公司，”莱弗尔斯继续说。

"And this is a modern tech company," Levels continued.

想象一下那些还在用PHP的传统公司。

"Imagine the traditional companies running PHP."

不过，即使那些公司没有这种无所谓的态度，这也仍然是个巨大的挑战。

Even for those companies not taking a blasé attitude, however, this represents an incredible challenge.

伊桑·莫利克再次写道：“我交谈过的公司都在为需要打补丁的速度而挣扎，因为AI防御系统在不可避免的AI攻击浪潮到来之前就发现了问题。”

Again, Ethan Mollick writes, "Companies I talk to are struggling with the rate of patches they need to do, thanks to AI defenses finding issues before the inevitable wave of incoming cyber attacks by other AIs.

现在消费者也会经历同样的事情，那就是需要频繁快速地更新所有东西。”

Now consumers are going to have the same experience of getting everything updated often and fast."

现在，有人对这次漏洞的披露方式提出了疑问。

Now, one question some had was around the disclosure method of this particular exploit.

Maple AI的首席技术官安东尼·罗宁写道：“如果他们研究这个PHP漏洞已经两个多月了，那为什么披露和部署策略这么糟糕？”

Maple AI CTO Anthony Ronning wrote, "If they've worked on the PHP vulnerability for over two months, then why has there been such a poor disclosure and rollout strategy?

这基本上就是一种‘24小时内你们全完了’的策略。”

It's basically been a strategy of, quote, 'In 24 hours, you're all screwed.'"

网络安全专家泰·莫纳汉写道：“这可能是个愚蠢的问题，但鉴于上周的发现以及12个多小时前就开始泄露，为什么PHP补丁没有在漏洞/概念验证完全泄露时立即发布？”

Cybersecurity professional Tay Monahan wrote, "This might be a stupid question, but given the discovery last week and that it started leaking 12-plus hours ago, why weren't the PHP patches ready to be dropped as soon as the vulnerability/POC fully leaked?

为什么大家都愿意再干等12个小时才打补丁呢？

Why is everyone okay arbitrarily waiting another 12 hours to patch?"

Christian the Pav写道：“我不是安全专家，但告诉黑客界PHP有个超级巨大的末日级漏洞，还给他们一天时间去破解并尽可能造成破坏，然后再发布修复，这听起来像是个坏主意。”

Christian the Pav writes, "I'm no security expert, but telling the hacking world there's a mega huge apocalyptic exploit for PHP and giving them a day to figure it out and do as much damage before they release the fix seems like a bad idea.

他们能不能先发布更新，过一天再解释是什么漏洞呢？

Couldn't they release the update, then after a day, explain what it is?"

根据那些比我懂得多得多的人的说法，我的理解是，补丁本身就会让漏洞变得显而易见。

Now, my understanding from reading accounts from people who, again, know a lot more than I do about this, is that the patch itself makes the exploit obvious.

所以他们的担心是，一旦发布补丁，实际上也就等于把漏洞公之于众了。

So the concern was that as soon as they released the patch, they were also essentially making the exploit known in the wild.

这就是为什么NSA提前警告大家要在披露之前把网站下线。

This is why the NSA warned everyone to take their sites down ahead of the disclosure.

Frederik Willems写道：“真正的黑客界不是问题所在。”

Frederik Willems wrote, "The real hacking world isn't the issue.

他们可能不会针对我们大多数人。

They probably won't target most of us.

发布这个漏洞或补丁，会引来一堆脚本小子，他们会扫描并攻破所有能找到的东西。

Releasing the exploit/patch unleashes a storm of script kiddies that will scan and breach everything they find.

他们试图让每个人在补丁发布的那一刻就准备好更新。

They tried to get everyone ready to update the moment it lands."

现在，在我们深入探讨更广泛的影响之前，有些人在感叹这个特定漏洞本身可能会留下的遗产。

Now, before we get into broader implications, some are lamenting what might be the legacy of this particular exploit itself.

Kenny Lauderdale写道：“PHP这事儿真让人沮丧，因为会有15年的老网站变成失落的媒体。”

Kenny Lauderdale writes, "This PHP stuff is so depressing because there's going to be 15 years' worth of old sites that are going to become lost media now."

网络安全新闻账号International Cyber Digest写道：“4chan已经变成只读，Something Awful也关闭了论坛，美国网站都在为PHP补丁做准备。”

Cybersecurity news reporting account, the International Cyber Digest, wrote, "4chan has gone read-only and Something Awful has shut its forums as American sites brace for the PHP patch.

书签服务Pinboard、游戏发行商Choice of Games、历史网站Damn Interesting，还有Explain XKCD维基也都下线了。”

Bookmarking service Pinboard, game publisher Choice of Games, history site Damn Interesting, and the Explain XKCD wiki are offline, too."

然后，为了说明这类攻击不会减少，昨天Railway也遭遇了一次重大漏洞攻击。

And then to add to the idea that these types of attacks were not going to be limited in their frequency, yesterday also saw Railway hit by a major exploit.

Railway是一个类似Vercel的AI托管服务。

Railway is an AI hosting service similar to Vercel.

在一份披露声明中，Railway写道：‘今天早上早些时候，我们检测到涉及存储客户环境变量系统的异常活动。',

In a disclosure post, Railway wrote, "Earlier this morning, Railway detected anomalous activity involving systems that store customer environment variables.

经过进一步调查，目前的情况表明，可能有未经授权的第三方访问了加密的环境变量，并获取了其中部分变量的解密内容。

After further investigation, our investigation to date indicates that an unauthorized third party may have accessed encrypted environment variables, and for some of them, obtained the decrypted contents.

环境变量可能包含诸如API密钥或数据库及注册表凭据等敏感信息。

Environment variables may contain secrets such as API keys or database and registry credentials.

我们认为这是近期针对平台公司的一系列攻击的一部分，其目的是破坏AI推理服务提供商。'

We believe this is part of a series of attacks recently targeting platform companies with the aim of compromising AI inference providers."

乍一看，这跟上周末的 OpenCode 漏洞非常像，攻击者再次获取了用户 API 密钥，从而能够窃取令牌并产生巨额账单。

At first glance, this looks very similar to the OpenCode exploit from last weekend, which again saw attackers gain access to user API keys, allowing the attackers to steal tokens and run up big bills.

跟 OpenCode 的情况一样，Railway 建议所有用户轮换 API 密钥，并重置平台上存储的所有环境变量。

Like the OpenCode situation, Railway is suggesting that all users should rotate API keys and roll all environmental variables stored on the platform.

SynthWave 的 Leo 写道："密钥正在被 actively 盗用，而且他们已禁止部署任何内容，所以无法使用新变量进行部署。"

Leo at SynthWave wrote, "Keys actively being drained, and they've disabled the ability to deploy anything, so can't deploy with new variables.

真是网络安全界的噩梦日，而且我担心情况只会越来越糟。"

Nightmare day for cyber, and it's only going to get worse from here, I fear."

OpenCode的Dax回应了Railway团队，并总结了这种新常态。

Dax from OpenCode responded to the Railway team and summed up the new normal.

他写道：“我相信你们都比我更擅长这个，但以下是我们看到的一些情况，希望对你们有帮助。”

He wrote, "I'm sure you're all better than me at this, but just some of what we've seen if it helps.

他们不会停手的。

They're not going to stop.

攻击会变得越来越复杂，他们会针对你系统的每一个部分以及你所有的员工。

It's going to get increasingly sophisticated, and they're going to go after every surface of your system and all of your employees.

任何你早已遗忘的旧项目都会被他们发现并用来对付你。

Any old projects you've long forgotten about will be found and used against you.

我们之所以还能撑住，是因为我们有全天候运作的代理。

The only reason we're surviving is because we have agents going 24/7.

我们仍然几乎没怎么睡觉。

We're still barely sleeping.

祝你们好运。

Good luck.

很抱歉你们要面对这种情况，但说真的，这是我们所有人共同面临的问题。

Sorry you're dealing with this, but really, it's a collective problem we're all facing together."

现在，如果这是新常态，那么这种常态可能会有非常不同的走向。

Now, if this is the new normal, there are very different paths that that normal can take.

一方面，前OpenAI员工Joshua Achaim提出了这样的观点，他写道：“感觉就像网络世界的2020年3月1日。

On the one end of the spectrum is what was suggested by former OpenAI staffer Joshua Achaim, who wrote, "Feels like March 1st, 2020 for cyber.

不正常。

Not normal.

有事情正在发生。

Something is happening.

几个月后互联网会被封锁吗？”

Internet lockdowns in a couple of months?"

另一方面，也有人认识到我们可能不得不以AI对抗AI。

On the flip side, there are those who are recognizing that we're probably going to have to fight AI with AI.

企业家John Ennis写道：“嗯，我们刚刚快速轮换了大量项目的所有密钥，我惊讶地发现，这在不久前绝对是一场噩梦。”

Entrepreneur John Ennis wrote, "Well, we just rotated every single key across a large number of projects very quickly, and I'm struck by the fact that this would've been an absolute nightmare not too long ago.

AI在网络安全方面的防御作用真的被低估了。

The defensive aspect of AI for cybersecurity is really underrated.

AI和核武器的类比有个不同点：你不能用核武器来阻止核武器，但你可以用AI来阻止AI的攻击。

Where the nuclear weapon analogy breaks down with AI is that you can't stop a nuclear weapon with another nuclear weapon, but you can stop an AI attack with an AI."

Box的Aaron Levie在这些观点上进一步发挥，写道：“网络安全将成为AI最深远的影响之一。

Box's Aaron Levie built on those themes, writing, "Cyber is going to be one of the most profound aspects of AI.

我们现在有了一支不知疲倦的队伍，可以搜寻、发现、修补或攻击任何已创建的系统或软件。

We now have a tireless workforce that can hunt down, discover, patch, or attack any system and software ever created.

我们会用智能体来识别现有系统和基础设施中长期存在的风险。

We will use agents to identify the longstanding risks in existing systems and infrastructure.

攻击者会用智能体来加速发现环境中的漏洞。

Attackers will use agents to accelerate the discovery of gaps in environments.

我们同样会用智能体来防御攻击，发现异常模式，更好地保护我们的数据等等。”

We will equally use agents to defend attacks and discover anomalous patterns to better protect our data and much more."

“我们正在进入一个新的网络时代，我们可能会在这个循环上使用比现在想象的多100倍的token，因为我们可以扫描所有日志，查看每一行代码，检查流量，并以以前无法理解的数据规模进行处理。”

"We're entering a new cyber regime, and we'll probably use 100x more tokens on this loop than we're imagining right now because we can scan every log, look at every line of code, look through traffic, and process data at a scale that wasn't comprehensible before.

企业现在比以往任何时候都更关心如何管理他们的网络、系统、软件、数据、身份、设备等等。

More than ever, enterprises are going to care deeply about how they manage their networks, systems, software, data, identity, devices, and more.

智能体将在很大程度上成为这些挑战的解决方案，但在这个过程中，它们会让每个人都保持警觉。

Agents will largely become the solution to these challenges, but they will keep everyone up in the process."

在另一篇文章中，Aaron写道：“如果你想知道AI会创造哪些工作，看看网络安全领域就够了。”

In another post, Aaron wrote, "If you want to know all the jobs that AI will produce, look no further than cybersecurity.

这将成为企业未来最大的投资领域之一。

This is going to become one of the biggest investment areas for enterprises going forward.

即使这些工作变得更加智能化和自动化，你仍然需要更多的人才来指导智能体处理最关键的工作、审查发现、修补系统、响应事件等等。

And even as that work becomes more agentic and autonomous, you will need more talent to point the agents at the most critical work, review findings, patch systems, respond to incidents, and more.

现在有很多机会，无论是保护数据和企业的系统，还是部署这些系统并保护组织的专业人士。

Lots of opportunity right now for the systems that protect data and enterprises, as well as the professionals that are deploying them and protecting organizations.

未来只会更加疯狂。”

It's only going to get more wild out there."

Embroidery的联合创始人Zach Corman用更生动的语言指出，唯一的出路就是直面问题。

Embroidery co-founder, Zach Corman, argues in more colorful language that the only way out is through.

他发帖说：“应对网络安全威胁只有一个答案，那就是修复你的系统。”

He posted, "There is only one answer to cybersecurity threats, fix your .

害怕威胁行为者使用人工智能吗？

Scared of threat actors using AI?

担心恶意智能体吗？

Worried about rogue agents?

觉得模型被摧毁很吓人？

Think obliterated models are spooky?

那就修复你的系统吧。

Fix your .

那些主张‘禁止AI模型、限制GPU访问’的人，应该先修复自己的系统。”

The 'ban AI models and regulate GPU access' crowd should instead fix their ."

然而，回到Aaron Levy刚才提到的技能缺口问题，有些人认为这种修复并不容易，也不是每个人都能做到的。

And yet, going back to the talent gap that Aaron Levy was just talking about, others think that that sort of fixing is not going to come easy and it's not going to be available for everyone.

Will Manidis写道：“如果今天的情况能预示未来的节奏，市场远远没有充分定价银行、医疗保健等领域为了抵御AI驱动的安全漏洞而必须进行合并以获取足够规模的速度。”

Will Manidis writes, "If today is any indication of the pace, markets are not nearly pricing in the rate at which banks/healthcare, et cetera, are going to have to merge to acquire enough scale to defend against AI-driven security vulnerabilities."

总结一下，Enclave的Tal Hoffman写道：“今天真是特别的一天。”

Summing it up, Enclave's Tal Hoffman writes, "What a day.

网络安全的未来是群体作战。

The future of cybersecurity is swarms.

一群攻击性智能体和一群防御性智能体，或者更准确地说，是舰队对抗群体。

Swarms of agents attacking and swarms of agents defending, or rather, fleets versus swarms.

有组织的防御方对抗廉价且不知疲倦的攻击方。

Governed agents on defense versus cheap and relentless agents on offense.

谁协调得更快，谁就获胜。”

Whoever coordinates faster wins."

在一篇题为《迈向保障互联网安全》的文章中，OpenAI总裁Greg Brockman写道，用他的话来说，“需要新的方法。”

In a post called "Towards Securing the Internet," OpenAI President, Greg Brockman, wrote about how, in his word, "new approaches will be needed."

他说：“我们正在积极思考这些问题，并期待社区的反馈，帮助我们共同应对随着模型改进而带来的互联网安全挑战。”

He said, "We are actively thinking about those and looking for feedback from the community to help us collectively rise to the challenge of improving the security of the internet as models improve.

随着漏洞发现速度加快，我们将需要新的防御解决方案，可能包括授权防御者，比如政府，扫描互联网寻找易受攻击的主机，并通知或极端情况下远程修补这些系统。”

As vulnerability discovery ramps, we will need new defensive solutions, potentially including authorized defenders, such as government, scanning the internet for vulnerable hosts and either notifying or, in extreme cases, patching those systems remotely."

在进一步讨论OpenAI的承诺之后，他总结道："我相信我们正进入网络安全的新阶段，模型能够发现具有互联网规模影响的漏洞。"

After discussing some more of OpenAI's commitments, he concludes, "I believe we're entering a new phase of cybersecurity, where the models are able to find vulnerabilities with internet-scale impact.

我们相信各地的防御者都能发挥作用，将模型应用于自身的防御，发现并修复共享基础设施中的问题，并分享最佳实践，最终让互联网更加安全。"

We believe that defenders everywhere have a part to play, applying models to their own defenses, finding and fixing issues in shared infrastructure, and sharing best practices that can ultimately make the internet more secure."

最乐观的看法是，最终软件将变得更加安全。

And the most optimistic take is that ultimately software will be more secure.

a16z的Martin Casado写道："从长远来看，AI将使软件更强大、更安全、更可靠。

a16z's Martin Casado wrote, "In the long run, AI is going to make software so much more capable and so much more secure and reliable.

但在此之前，情况会变得糟糕得多。"

But before that, things are going to get much, much worse."

长期听众会听到我说过，在AI的许多方面，我是一个坚定的长期乐观主义者，但长期乐观并不否认短期甚至中期转型可能带来的混乱、痛苦和煎熬。

Longtime listeners will have heard me say before that across so many aspects of AI, I am an incredible long-term optimist, and yet long-term optimism does not deny the potential chaos, pain, and agony of short or even medium-term transitions.

我相信工作领域是这样，网络安全领域更是如此，而我们现在正身处其中。

I believe that's the case for jobs, and it is certainly the case for cybersecurity, and boy, are we in it now.

当然，随着事态发展，我会随时向你们通报最新情况。

I will, of course, keep you informed as things evolve.

不过，今天就到这里，这是今天的AI每日简报。

For now, however, that is going to do it for today's AI Daily Brief.

一如既往，感谢你的收听或观看，下次再见，祝好。

Thanks as always for listening or watching, and until next time, peace.