本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
第一次，安全不再是刹车，而是油门。
For the first time, security is the gas pedal, not the brake.
你怎么描述你看到的AI对攻击者带来的影响？
How would you describe what you're seeing AI doing for adversaries?
新的顶级捕食者是智能体状态。
The new apex predator is the agent state.
几年前，只有国家才能拥有的能力，现在一个黑客行动主义者就能拥有。
The same level of capability that a nation state would have just a few years ago, essentially, a hacktivist would have today.
这已经彻底改变了整体的威胁环境。
That has dramatically changed the overall threat environment.
大家要记住的是，它……
What people have to keep in mind is it...
它不再是恶意软件了。
It's not malware anymore.
它是提示词。
It's prompts.
他们基本上可以设计出一些提示，从而创造出完全独特的攻击方向、信息收集方式以及转向点的特征。
They can basically engineer prompts that would create a totally unique signature of what direction they're gonna go in, how they gather information, where they pivot.
每个人都来找CrowdStrike和其他公司，问：我们该如何确保AI的安全，以便能更快地推进AI发展？
Everyone's coming to CrowdStrike and others going, how do you make AI secure so we can go faster in AI?
今天，只有更好的AI才能对抗AI。
The only way to fight AI today is with better AI.
防守方最了解自己的系统。
The defender knows their own systems.
防守方掌握着架构图。
The defender has the architectural diagram.
防守方拥有代码。
The defender has the code.
你怎样才能把这些全部整合起来，从而大幅提高自己战胜进攻性AI的概率呢？
How do you put all of that into something that you can dramatically increase your probability of outsmarting the offensive AI?
嘿，听众朋友们，大家好！
Hi, listeners.
欢迎回到 No Priors。
Welcome back to No Priors.
今天，我和埃拉德请到了乔治·库尔茨。
Today, Elad and I are here with George Kurtz.
乔治是全球网络安全事件发生后第一个被求助的人，他是一位传奇的网络安全捍卫者，也是 Foundstone 和 CrowdStrike 的创始人兼 CEO。
George is the world's first call when there's an incident, a legendary cyber defender, the founder and CEO of Foundstone, and then CrowdStrike.
我们谈谈网络安全领域的现状，如何用 AI 对抗 AI，他们内部的研发项目 Safemind，以及互联网在下一阶段是否前景光明。
We talk about what's happening in cybersecurity, fighting AI with AI, their internal research effort, Safemind, and whether the odds are good for the Internet in the next era.
欢迎回来。
Welcome back.
乔治，非常感谢你接受我们的邀请。
George, thanks so much for doing this.
我很荣幸能来。
It's a pleasure to be here.
现在科技正处于巨变时期，智能体正在接管企业运营。
So it is a sea change period in technology, and agents are taking over the enterprise.
CrowdStrike在这方面走在了前沿。
CrowdStrike is on the forefront of that.
你现在观察到的情况能跟我们说说吗？
What can you tell us about what you're you're seeing in in the wild right now?
我记得你提过，用户环境里实际在行动的智能体数量比他们意识到的要多出几个数量级。
I think you have said that there are, you know, two, three magnitudes more agents out there taking action in in people's estates than they're aware of.
在企业领域，我们看到的景象真是太惊人了。
It's been incredible in the enterprise what we've seen.
只要看看前沿实验室的发展轨迹和它们自己的收入，就能发现智能体的数量有多庞大，比如Claude Code或Codex到处冒出来。
If you follow the trajectory of the frontier labs and you just look at their own revenue, you can see just how many more agents are out there, whether it's Claude Code or Codex popping up all over the all over the environment.
当然，还有很多其他类型的智能体，但用这个作为参照就足够说明了。
Obviously, there's many other type of agents that are out there, but again, just using that as a proxy.
而且你知道我们发现什么了吗？它们无处不在，但这些企业的问题就是，他们根本不知道这些智能体在哪些地方冒出来的。
And you know what we've seen is that they're popping up everywhere and part of the problem in these enterprises is they have no idea where they're popping up.
影子AI已经成为很多公司的一大头疼问题。
Shadow AI has become a huge issue for many companies.
我知道我们接下来要聊的是这些代理都在干什么，以及发生的那些不好的事情，但总的来说，让企业了解他们的AI在哪里运行、谁在做什么、更重要的是，他们实际上是以什么身份在操作，这一直是行业里的一大挑战，也是我们CrowdStrike在为客户破解这个难题时的重点。
I know we're going to talk about what the agents are doing and all the bad sort of things that are happening, but just in general, having organizations understand where their AI is taking place and who's doing what, and more importantly, what identity are they actually operating under has been a big challenge in the industry and a big focus for us at CrowdStrike in in trying to crack the code for our customers.
你如何看待AI在帮助对手方面所起的作用？
How would you describe what you're seeing AI doing for adversaries?
你之前把它描述成压平了对手的金字塔。
You've described it as flattening the the adversary pyramid before.
说到对手，我称之为痛苦金字塔。
When I think about the the adversaries, I I call it the pyramid of pain.
你可以想象一下金字塔的底部，底座其实就是我们所说的黑客行动主义者。
So if you just imagine the bottom part of the pyramid, the base of the pyramid is really what we call a hacktivist.
这一层次的人技术水平最低，但人数最多。
So you've got the least level of sophistication, but the greatest number of adversaries.
在金字塔的中间层，是网络犯罪团伙。
Then in the middle band of the pyramid you've got e crime.
他们更老练，人数也比黑客行动主义者少一些。
They're more sophisticated and a few less than all the hacktivists.
然后金字塔顶端就是国家级行为者，对吧？
Then at the top of the pyramid you've got nation state actors, right?
他们是最精通技术的，但相对而言人数最少。
So the most sophisticated, but the least number relatively speaking.
但现在这一切完全被颠覆了，因为我们看到的新顶级掠食者不再是国家，而是智能体。智能体已经改变了游戏规则。
Well, that has been totally just blown up because really now what we're seeing is the new apex predator is the agent state, not the nation state, it's the agent state and what the agent state has done.
如果我们想到那些被削弱了防护的模型，基本上没有任何真正的护栏，它就是把所有这些非常复杂的话题、技术和成本，全部压平了整个金字塔。
If we think about the models that are out there that are ablated so there's no real guardrails, it's essentially taken all of these very complex topics and sort of techniques and cost and now flatten the entire pyramid.
所以，几年前只有国家级行为者才有的同等能力，
So the same level of sort of capability that a nation state would have just a few years ago.
如今一个黑客活动分子就能拥有，这彻底改变了整体的威胁环境。
Essentially, a hacktivist would have today, and that has dramatically changed the overall threat environment.
那我们看到了什么？
So what are we seeing?
我们看到更多的漏洞被发现，攻击速度更快，而且攻击是一波接一波地来。
We're seeing more vulnerabilities found and we're seeing faster attacks, and when the attacks come, it's just, you know, it's just one after another.
你可以看到的。
You can see it.
其实现在还挺明显的，是不是AI攻击，因为你能看到它们迭代的速度有多快。
It's actually right now it's pretty clear whether it's an AI attack because you can see how fast they're actually iterating.
但总有一天，它们会放慢节奏，变得隐蔽，那时候就难分辨哪些是AI干的，哪些不是了。
But at some point, they're gonna introduce that being low and slow, it's gonna be more difficult to figure out what's AI and what isn't.
但这就是我们现在看到的情况，而且我觉得组织面临的一个真正的挑战就是攻击接连不断，而且攻击的复杂程度也大大提高了。
But that's what we're seeing, and I think that's one of the real challenges that organizations are facing is just the barrage of attacks and the level of sophistication has dramatically increased.
最终，如果你给一个AI代理一个任务，给它足够的时间，它要么想办法攻进去，要么就死掉，所以它会想方设法突破，而且很多时候它都能成功。
And ultimately, if you know, if you give an agent a task with enough time and, you know, the agent either needs to get in or dies, it's gonna try to figure out a way to get in, and a lot of times it's gonna have success.
你看到这些对抗性攻击的方式有什么变化吗？
Are you seeing a shift in terms of how these adversarial attacks are being made?
我知道我们之前聊过。
I know we were talking earlier.
我们聊了一点关于开源模型、闭源模型，还有开放权重模型的话题。
We talked a little about open versus closed source models or open weight models.
你有没有发现大多数攻击用的都是开放权重模型？
Are you seeing a shift where most of it is open weight models?
还是说人们在用顶尖实验室的最新技术，是两者混合的情况？
Is it a mixture in terms of people using state of the art labs and capabilities?
比如说，现在人们到底是用什么来做这些攻击的？
Like what what are people actually using to do these attacks at this point?
我们看到很多攻击确实是在开放权重模型上进行的。
We're we're seeing a lot of it on the open weight models, right?
你只要看看最新一代前沿模型的防护措施。
If you look at, I mean, just the guardrails on the latest generation of any of the frontier models.
说实话，连做一个包含网络安全内容的PPT都很难。
To be candid, you can barely create a PowerPoint that has cyber security in it.
它直接就会拒绝你，把你降级到低一级的模型。
I mean, it literally rejects you and bumps you down to a lower model.
所以我们看到的情况是，比如GLM、Quinn，或者其他那些模型，它们在处理网络安全方面非常有效，而且很多模型要么是经过消融处理的，要么就是专门为网络安全而建的。
So what we've seen is, you know, again, if you look at, GLM, if you look at, you know, Quinn, if you look at any other sort of models that are out there, they've been very effective in being able to deal with cybersecurity, and many of the models are either ablated or specifically built just for cyber.
可能是用于红队测试，这类攻击你实际上可以直接运行。
It might be for red teaming, and those are the type of attacks where you can actually just run these.
你可以直接跑起来。
I mean, you can run them on.
我家有两台 Spark DGX，你知道的，所有模型都在跑，想干什么就干什么。
I've got two spark DGXs at home and you know I've got all the models running and you know whatever I want to do, I can do.
你想做勒索软件？
You want to create ransomware?
没问题。
No problem.
想做个勒索软件工具包？
You want to create a ransomware kit?
也没问题。
No issue.
想部署？
You want to deploy?
你想迭代吗？
You want to iterate?
我是说，所有东西都给你准备好了，你知道，这跟直接在笔记本上跑就差一步了。
I mean it's all there for you and you know that is like one step removed from just running it all on your laptop.
我们再想想现在Mac的情况。
And we think about where Mac is today.
就算是即将推出的新款Mac，比如Studio系列，它们也会成为非常强大的AI机器。
Even the new Macs that are coming out, The studios, they're going be incredible AI machines.
所以最前沿的就是那些没有护栏就能运行的东西，而前沿模型在比如需要PowerShell脚本之类的方面也挺有帮助的。
So the tip of the spear is gonna be what you can run without the guardrails and the frontier models are sort of helpful in terms of if you needed PowerShell scripts or those sort of things.
但当然，现在用开放权重模型就能搞定很多这类事。
But certainly you can do a lot of it with the open weight models right now.
那么，这些攻击有没有一个共同的地理区域呢？
And is there a common geography?
比如，是不是大部分攻击都来自同一个地区？
Like are most of the attacks coming from a single region?
有没有……
Are there...
有没有一种常见的攻击类型？
Is there a common attack type?
主要是勒索软件吗？
Is that mainly ransomware?
还是别的什么？
Is it something else?
是钓鱼活动吗？
Is it phishing campaigns?
我有点好奇，随着AI工具的普及，攻击的类型有没有发生变化？
I'm little bit curious, like has the mix shifted as you've kind of seen the adoption of AI tools?
我们看待这个问题的角度可能有点不同，因为我们是从对手的角度来看的。
We sort of look at it, maybe just reframe the question a little bit, because we look at it from an adversary perspective.
是国家级的行为吗？
You know, is it a nation state?
是网络犯罪吗？
Is it e crime?
对于国家行为体，我们起了这些名字，比如熊、熊猫，取决于它是来自俄罗斯还是伊朗，还有小猫之类的。
Nation state, we sort of, you know, we've come up with these names of bears and pandas, depending if they're Russia or it could be from Iran, kittens, and other things.
所以我们看到了很多攻击。
So, but we've seen a lot of attacks.
我们看到朝鲜在这些攻击上非常活跃。
We see North Korea being very busy with these sort of attacks.
他们需要，怎么说呢，把钱引入他们的经济，所以他们一直在忙着窃取信息。
They've got to, shall we say, get money into their economy, so they've been very busy on, you know, stealing information.
很多都是关于勒索软件和收钱的。
A lot of it around ransomware and getting paid.
但总的来说，我们看到很多……
But yeah, I think overall what we've seen is a lot of...
显然，我们看到中国方面有很多使用AI的活动，而且在网络犯罪领域也看到了，当有人进入系统时，我们会追踪突破时间。
We've seen obviously out of China, a lot of activity using AI, and we've also seen it in the e crime world where when someone gets onto a system, we track breakout time.
也就是说，如果他们进入了你的系统，他们实际需要多久才能突破并转移到另一个系统？
That is, if they got onto your system, how long would it take them to actually break out and pivot and move to a different system?
一般来说，他们最初进入的系统并不是最关键的。
Generally, the system they first land on is not the most interesting.
有可能，但它不是最有价值的目标。
It could be, but it's not the most interesting.
最重要的数据可能在别的地方。
The crown jewels might be somewhere else.
所以平均来说，过去这需要几天甚至几小时，现在只需要几分钟。
So on average, that used to be days and hours, and now it's minutes.
平均时间是二十九分钟，但我们见过最快二十七秒就完成的。
On average, it's twenty nine minutes, but we've seen it as quick as twenty seven seconds.
现在本质上我们关注的是推理速度。
And essentially, you know, now what we're looking at is the speed of inference.
所以一旦他们进入机器，就能看到它很快地反复尝试。
So as soon as they get on the machine, can just see it iterating very quickly.
大家要记住的是，这已经不是恶意软件了。
And what people have to keep in mind is it's not malware anymore.
现在是提示词。
It's prompts.
所以，他们基本上可以设计出提示词，这些提示词能创造出完全独特的攻击方向、信息收集方式以及横向移动路径的签名，这使得我们更难搞清楚这些。
So they can basically engineer prompts that would uniquely create a totally unique signature of what direction they're going to go in, how they gather information, where they pivot, which makes it more challenging to figure all that out.
但这就是我们现在看到的，而且这种情况已经遍布各个层面，无论是国家级行为者还是网络犯罪团伙，都是如此。
But that's what we're seeing right now, and it's across the board from the nation state and e crime actors for sure.
你从Foundstone、CrowdStrike开始就一直担任安全领域的领导者……
You have been a leader in security since, you know, Foundstone, CrowdStrike...
创立CrowdStrike。
The founding of CrowdStrike.
你会说他们……
Would you say that they're...
就是说，你如何将当前攻击者和防御者能力的提升置于背景中看待？
Like, how do you put the current period of increase in capability, like attacker capability and perhaps defender capability in context?
我们现在是不是处于一个攻方更占优势的世界？
Are we in a more adversary dominant world right now?
你是怎么看待这个整体问题的？
Like, how do you think about the problem overall?
嗯，我觉得现在正是科技领域最激动人心的时代之一。
Well, I think it's one of the most exciting times to be in technology.
真的，我已经干这行三十多年了。
I mean, really, I've been doing this thirty plus years.
我总爱说，安全的演进跟技术曲线是同步的。
And, you know, what I always like to say is that security parallels the slope of the technology curve.
我刚入行的时候，也就是有个思科路由器，可能连网站都刚刚起步，对吧？
When I first got into security, it was a Cisco router and like maybe a website barely, right?
那时候真的没什么防火墙。
There really weren't firewalls.
有的只是那些工具包，像TIS toolkit之类的，我这么说可能暴露年纪了，但你懂的，那会儿真没什么东西。
There were these tool kits, TIS toolkit and these other things probably dating myself, but you know there wasn't much there.
现在你想一下我们今天的复杂程度，有云，还有AI周围的各种技术，我不打算一一列举，但今天比那时候复杂多了。
Now if you think about the complexity of where we are today with with cloud, with all of the other technologies around AI, I'm not going go through all of them, but it's a lot more complex today than it was way back then.
所以你需要所有这些小公司，包括大公司，能够保护所有这些技术。
So you need all these little companies and including big companies that are able to secure all these technologies.
所以当我们想到今天的状况，这实际上是最令人兴奋的，但同时也创造了最大的机会，因为对手通常能找到漏洞或缝隙。
So when we think about where we are today, it actually is the most exciting, but it also creates the biggest opportunity because the adversaries generally find the gaps or the seams.
他们会找到新技术，可能这些技术没有很好的保护，他们会利用像NPM包这样的东西，在以前，你也知道，企业界的美国人不会随便下载任何东西就跑到电脑上运行。
So they'll find new technologies that maybe that they don't have good protection, that they're taking advantage of things like NPM packages, In the old world, as you as you very well know, no one in corporate America would just download anything and run it on their computer.
那必须经过审查和扫描等等这些程序。
It had to be vetted and scanned and all these sort of things.
现在我们释放一个代理。
Now we unleash an agent.
我们给它一个身份，本质上是我们的身份或者更糟，然后我们释放它，说，你想做什么就做什么。
We give it an identity, which is essentially our own or worse, and then we unleash it, and we say, do whatever you want to do.
顺便说一句，你想下载什么包就下载什么包。
And by the way, download whatever package you want to download.
很明显，这类事情会造成巨大的安全风险。
And obviously, these are the kind of things that cause a tremendous amount of exposure.
所以我们正在逐渐习惯这一点。
So we're sort of getting to grips
而且没有人会去检查生成的代码。
And nobody's reading any of the production code as it gets.
对，确实没有人检查。
No, they're not.
你过一阵子看到这些东西，真的会让人摸不着头脑。
I mean, you know, you see some of these things in a while, then you're scratching your head.
但这也不应该让你意外。
You shouldn't be surprised.
就好比说，给它一个任务，然后过几天再回来，希望一切都正常。
I mean, it was like, let's give it a task, and we'll come back in a few days, and we'll hope it all worked.
根本没人盯着它。
Nobody's watching it.
没人知道它在做什么，这确实是个问题。
Nobody understands what it's doing, and it's problematic.
所以你知道，我们可以。
So you know we can.
我们可以谈谈这些智能体的可解释性，这将会非常重要，但目前它就像一个黑盒，没人知道里面在发生什么。
We can talk about sort of the explainability of what these agents are doing, which is going to be really important, but it right now it's like a black box and nobody knows what's going on.
他们希望你知道，有什么东西在运作，然后从另一端出来，但现在是投身科技领域的好时机。
They hope you know something happens and pops out on the on the other side of it, But it's a great time to be in technology.
人工智能将从根本上改变世界，前沿实验室在做很棒的事情，但最终还是要靠安全来支撑这些技术。
AI is fundamentally going to change the world and the frontier labs are doing amazing things, but ultimately it's going to come down to security enabling these sort of technologies.
安全第一次成为油门，而不是刹车。
For the first time, security is the gas pedal, not the brake.
这就是我们所看到的。
And that's what we're seeing.
你是从什么时候意识到这个拐点正在发生，至少是在安全领域？
What is the moment that you realized this inflection was happening, at least in the security world?
我觉得对很多人来说，Hugging Face 和 OpenAI 那件事是个大转折点，大家这才意识到能力、技术以及可能发生的事情都有了巨大的变化。
And I think for a lot of people, the Hugging Face OpenAI issue was a big moment where people realized there's a big c shift in terms of capabilities and technology and what could happen.
而且就像你说的，这些事情有没有被监控等等。
And to your point, are things being monitored, etcetera?
那对你来说，那是不是一个关键时刻，还有哪些
Was was that a seminal moment for you, what are some
关键时刻呢？
of the seminal moments?
让我先问你一个问题。
Let me ask you a question first.
那从 Hugging Face 这件事里，你的最大收获是什么？
So what was your big takeaway from Hugging Face?
那就是，任何运行前沿模型的公司都需要注意他们的环境质量，并且要有非常好的持续生产监控。
That anybody running frontier models needs to be careful of their environment quality and have very good ongoing production monitoring.
这就是收获。
That is the takeaway.
但我想很多人也会说，这是第一次真正的端到端代理攻击。
But I think what a lot of people would say as well, this was the the first agentic end to end attack.
没错，你看看这些家伙在做什么，人们说他们会隐藏自己，试图撑到下一代之类的。
All true, and you kind of look at, you know, what these things were doing, and you know, people talk about them, you know, hiding themselves and you know, trying to basically survive to the next generation and those sort of things.
确实是。
All true.
我最大的收获是，防御方没有攻击方那样的能力。
My biggest takeaway was that the defenders didn't have the same capabilities as the offense.
好吧，现在你会看到这个，说安全气囊被关了，因为这是一个实验，代理和护栏都被解除了。
Okay, now you would look at this and say the airbags were off because this was an experiment where the agents, you know, the guardrails were down.
安全气囊被关了，最后他们还是找到了逃出沙盒的办法。
The airbags were off, and ultimately they found a way out of the sandbox.
但现实是，Hugging Face 做的第一件事就是去找那些前沿模型，然后说，这玩意儿是真的吗？
But the reality was Hugging Face, you know, when the first thing they did was they went to the frontier models and they said, what like, is this thing real?
到底怎么回事？
What's going on?
然后他们被拒绝了，对吧？
And they got refused, right?
因为那边有安全措施。
Because the airbags were on over there.
所以他们该怎么办？
So what do they have to do?
他们实际上不得不求助于开放权重模型来筛选所有这些数据并搞清楚。
They actually had to go to open weight models to kind of sift through all this data and figure it out.
所以对我来说很重要的一点是，确保防御方拥有与攻击方同等级的最前沿防御性AI。
So a big piece for me is to basically make sure that defenders have the same level of frontier defensive AI as what the offense is using.
这是我的一大收获，也是CrowdStrike正在努力的方向之一。
And that was a big takeaway for me and part of the things that CrowdStrike is working.
在你提到的背景下，这难不难？因为那些顶尖实验室的模型防护太强了，以至于你可以说，并不是每个企业都能获得同样的防御能力。
And is that hard in the context that you mentioned of the guardrails being so strong on the state of the art labs models that to some extent you could argue there aren't those same defensive capabilities available to every enterprise?
换句话说，他们不得不使用开放权重模型的事实表明，目前实际上无法用那些实验室的模型做到这一点。
In other That's words, fact that they had to go to open weight models suggests that you can't actually do it currently with the labs.
那么问题来了，你们怎么采用这个呢？
And so the question then, how do you adopt that?
比如你想用前沿模型做一份关于网络安全的PPT，它直接拒绝你。
I mean, like you try to make a PowerPoint in with a frontier model around cybersecurity and it refuses you.
这就是我们创建SafeMind以及红队和蓝队模型的部分原因，因为我们想给客户提供前沿级别的防御能力，让他们不用面对各种拒绝就能操作。
So, you know, this is part of why we created SafeMind and red and blue models just for that reason, because we wanted to give frontier caliber defenses to our customers that allow them to operate without sort of all these sort of refusals.
我认为这将是未来许多公司的一个重要元素。
And I think that's gonna be an important element going forward for for many companies.
我觉得...
I think...
我想回到攻防的对比上，似乎每次技术变革时，进攻能力总是稍稍领先，因为做事方式出现大变化，有几个原因。
I guess back to the offense versus defense side, it seems like with every technology shift, offensive capabilities are always a little bit ahead when you have a big change in how things are done for a few reasons.
其中之一是，进攻方可以非常直接。
I mean, one of them is just with offense, it can be very directive.
我需要去做某件事，而防守方需要覆盖一个广泛的面。
I need to go do x, while with defense, it's a broad surface area you need to cover.
所以在某种程度上，你需要覆盖更多的东西。
So to some extent, you need to cover more stuff.
其次，就是技术的采用和扩散。
And then secondly, it's just technology adoption and diffusion.
你是一个攻击者……
You are one attack...
你是一群攻击者。
You are a set of attackers.
你知道你要针对的是什么，而企业防御面对的却是分散的生态系统。
You know what you're going after versus the diffuse ecosystem of enterprises that you need to defend against.
CrowdStrike是如何考虑确保这些防御能力被广泛可用的呢？
How does CrowdStrike think about ensuring that those defensive capabilities are very broadly available?
嗯，这是我们关注的重点之一。
Well, this is one of the things that we focused on.
我们在“网络超级智能实验室”这个词流行起来之前就创建了它，我可能会这么说。
We created our own cyber superintelligence labs before the term was popular, I might say.
所以我想，这个我们算是做得很不错了。
So I guess we we kind of nailed that one.
但本质上，我们做的是创建了一个自己的前沿实验室，专注于安全模型，包括红队、蓝队和整体框架。
But essentially what we've done is we've created our own basically frontier type lab that is totally focused on security models, both red and blue and the harness.
我们把它叫做Safe Mine系统，但我们有红队Tempest和蓝队Solano。
And we call it safe mine as a system, but we've got red Tempest and blue Solano.
这两个就是我们的模型。
Those are the two models.
我们跟英伟达在这方面合作了很多，我们基本上利用了……
Now we worked extensively with Nvidia on this and we basically leverage.
我们利用了多种模型类型，但Nemo Tron是我们蓝队主要用到的模型之一。
We leverage a few model types, but Nemo Tron is one of the main models that we leverage on on the blue side.
当我们坐下来跟他谈，我也和Jensen坐下来讨论的时候，过去六个月我们一直在推进这个项目，我们仔细研究了它的运作方式，而本质上正如你所说，防御方似乎总是慢一步。
And when we sat down with him and I sat down with Jensen, we were working on it for the last six months and we kind of went through how this would work and essentially to your point like you know the defense always seems to be one step behind.
我们做的主要工作之一，就是不断迭代红队和蓝队。
The big part of what we're doing is we're iterating the red and the blue.
对于我们的客户，我们能看到他们环境的样子。
For our customers we have a view of what their environment looks like.
然后，这正是Jensen坚持的，说你们必须有一个数字孪生。
And then, and this was Jensen sort of insisting like you have to have a digital twin.
你看其他行业，他们都有数字孪生。
If you look at other industries, they have digital twins.
他就说，你们安全领域也必须有一个。
And he goes, you have to have one security.
所以我们建了一个数字孪生，我们可以把关于客户环境的所有信息都放进去。
So we built a digital twin where we can take all of the information that we know about a customer's environment.
我们可以基本上在虚拟世界里重建它，然后我们可以迭代模型，让模型循环运行，红色模型不断尝试攻入，直到它的回合用尽，因为蓝色Solano已经看穿了红色的行动。
We can basically recreate that in a virtual world and then we can iterate the models and loop the models where red continually is focused on getting in until it runs out of turns because Blue Solano has actually figured out what RED was doing.
我认为这就是我们跟上威胁的方式，因为红色模型在不断迭代，所以你必须要有这样一个循环机制。
And that's really the way I think we have to keep up with the threats because RED is constantly iterating, and you're gonna have to have this loop in place.
有道理。
That makes sense.
我是说，即使从传统的网络安全角度来看，了解你的环境是你作为防御者仅有的优势之一，而CrowdStrike也一直在引领，比如共享情报和理解对手的能力。
I I mean, even just from a classic cybersecurity perspective, like knowing your environment is one of the only advantages that you have as a defender, as well as as as CrowdStrike has pioneered, like, you know, the ability to share intelligence and understanding of the adversary.
实际上，我觉得有必要退一步，整体描述一下CrowdStrike目前在研发方面的主要产品和投资。
Actually, I I think it'd be useful to just, like, zoom out and describe, like, you know, the the key offerings and investments on the R and D side that CrowdStrike is making today?
因为我觉得它是个巨大的网络安全玩家，但可能大家还停留在端点保护和了解对手这个老观念上。
Because I I think like it's a it's a massive cyber player, but I think the the root of like endpoint protection and understanding the adversary might be an outdated view.
我觉得没错。
I think that's right.
过去大概四五年，我们真的已经扩展到了更广泛的平台能力。
You know, we we've really, you know, probably over the last, you know, four or five years, I guess, really moved into, you know, much broader platform capabilities.
我们现在有33个模块，你知道吗，刚开始创办公司时只有一个，现在有33个了。
We've got 33 modules today, and you know, ultimately, when I started the company, had one module, and now we've got 33.
我们不仅构建了端点安全的能力，还覆盖了云、IT、身份以及许多其他领域。
We've built capabilities not only across endpoint, but around cloud, IT, identity, and many different areas that are out there.
所以，看看我们在哪里投入了很多，以及我们在最新的用户大会Falcon上谈到的，主要是我们所说的AIDR。
So if you look at where we've made a lot of investments and some of the things that we just talked about at our latest user conference, Falcon, it was around what we call AIDR.
所以我们开创了EDR，也就是端点检测与响应，早期很多时候就是一台电脑加一个人。
So we pioneered EDR, Endpoint Detection Response, which a lot of it in the early days was a computer and a human.
然后你知道，我们显然围绕这个加了很多别的东西，比如防护之类的。
And then you know we've added obviously many more things around that and prevention and all those sort of things.
但当你看看一个代理在做什么，它本质上就像个超人。
But when you look at what an agent is doing, it's essentially a superhuman.
它有身份。
It has an identity.
它能访问计算资源。
It has access to compute.
它能在你的电脑上运行，或者跑在容器里。
It runs on your computer or it can run-in a container.
它还能访问数据和网络能力，对吧？
It has access to data as networking capabilities, right?
所以我们想做的就是先找出所有这些杂乱的和影子AI，因为没人知道它们在哪，就像你刚才说的。
So what we wanted to be able to do is to first find all of this rogue and shadow AI because nobody knew where it was like back to your point.
你得了解你的环境是什么样的。
You got to know what your environment looks like.
其次，我们可以从智能体诞生那一刻起，就对其行为进行监控。
And then two, we can instrument from the birth of an agent what that agent does.
所以如果你在写代码，你启动一个智能体，它会有子智能体。
So if you're in, you know, clogged clogged code, you start an agent has sub agents.
它会有工具调用。
It has tool calls.
它会连接到MCP服务器，获取数据，做一些事情。
It you know goes to an MCP server, pulls data, whatever it's doing.
我们实际上可以告诉你整个事件链，这非常重要，然后能够围绕这些添加策略和阻止规则。
Can actually tell you the entire chain of events, and that's incredibly important and then be able to add policy and blocking rules around those.
所以我们在那个领域花了很多时间，从商业角度来看，如果你看看现在的统计数据，平均来说，虽然我觉得这个数字很低，但据说一个员工可以有90个不同的智能体。
So we've spent a lot of time in that area, and I think from a business perspective, if you look at the stats right now on average today, which I actually think is low, but they say you can have 90 different agents, you know, as an employee.
那么，这对我们来说可能比EDR有更大的机会，所以这是我们一直关注的领域之一。
Well, that can be a much bigger opportunity than just EDR for us, so that's one area that we've been focused on.
我要说的第二个方面，就是我之前提到的，建立超级智能实验室，打造我们自己的前沿模型。
The second area then, as I mentioned, is creating this super intelligence lab and creating our own frontier models.
再说我们的模型，我们是从现有的模型开始的，比如Nemotron这个类别的，主要是Nemotron三系列的。
Again, our models, we start with existing models like a Nemotron class, you know, mostly the three class of Nemotron.
我们帮它们微调一些东西，满足即将推出的那个产品的需求，然后我们对它们进行后续训练，利用我们过去十五年积累的宝贵数据，来获得最好的效果。
We're helping them sort of fine tune some of the things we need for the force that's coming out, but then you know we kind of, you know, post train them and using all the valuable data that we have over the last fifteen years to get the best outcome.
最后我再补充一点，然后我就先停一下。
And I'll say one more thing, and I'll pause.
那个架架构带来了巨大的不同。
The harness makes a huge difference.
几乎我们选的任何一个模型，不管它是前沿模型，还是开源的现成权重模型，靠我们配套的这个架构，我们都能以更低的成本得到更好的结果。
Like almost any model we take, whether it's frontier, whether it's just an off the shelf open weight model, we can get much better results at a way lower cost because of the harness that we created that goes with it.
以上就是我们一些大的投入方向。
So those are some of the big investments.
身份认证领域还有更多的进展。
You know, there's more around identity.
我很乐意详细讲讲这些，但先让我停一下，看看你们对此有没有什么其他评论或想法。
I'm happy to cover any of those, but let me pause there and see if had any any other comments or thoughts on those.
这本来是我要问的关于你们实验室投资的大问题之一。我觉得对于任何模型训练工作，都会有人问一个合理的问题：为什么不直接依赖前沿模型？
That was going be one of my big questions about your labs investment here, which is I think a reasonable question to ask about any model training effort is like, why isn't that just a frontier?
你知道，你们和OpenAI、Anthropic合作，对吧？
You you know, you work with open AI and Anthropic Yeah.
比如能力方面。
Like capability.
这些实验室当然也明白安全的重要性，他们有针对经过审查的防御者的项目，比如Daybreak或者某种Mythos权限。
These these labs, of course, they realize the importance of security, and then they have these programs for, like, vetted defenders, be it, like, Daybreak or sort of mythos access.
为什么这还不够？
Why is that insufficient?
你们有基于工具链、你们内部的理解和数据的能力优势。
I mean, you have you have a capability argument based on, you know, harness your your understanding internally data.
除此之外，你们认为对于独立安全供应商来说，还有哪些结构性因素很重要？
Is there anything else structurally that you think is important for an independent security vendor here?
对，我觉得这里面有几个方面。
Yeah, I think there's a few things there.
首先，这些实验室做的事情太厉害了。
So first, what the labs are creating is incredible.
他们从哪儿开始，又做出了什么样的成果，简直不可思议，对吧？
I mean, where they've come from, what they created, it's unbelievable, right?
我们也想把这种能力带给我们的客户。
And we also want to make that available to our customers.
所以总体思路是，如果你有一个安全可靠的想法，想用前沿模型，那太好了。
So the general theory is if you have a safe mind and you want to leverage a frontier model, great.
我们也能帮你利用这种模型。
We can we can leverage that.
如果你想用我们的模型，那也非常棒。
If you want to leverage our models, fantastic.
我们的模型会便宜很多。
Our models will be significantly cheaper.
举个例子，就拿一般的安全修复来说，如果你用最高端的网络领域前沿模型，可能大概要花十美元。而我们的只要三美元左右。所以成本差异非常大，这一点很重要。
So just to give you an example, so for general remediation, if you use like the highest caliber frontier cyber model, it probably costs you about $10 Ours is like $03 So there is a huge cost difference, is important.
不过再看看我们的模型，因为我们在自己的数据上做了后期训练，一个巨大的差异就是我们拥有这些数据。
But then when you look at, you know, our model because of the way we've post trained it on our data, a huge difference is we have the data.
就像不存在什么CrowdStrike安全数据的Reddit一样。
Like there is no Reddit of CrowdStrike security data.
你可以生成合成数据，但那真的比不上我们手里的真实数据。
You can create synthetic data, but it's really not the same as what we have.
我们有十五年的积累。
We got fifteen years.
嗯，可能稍微少一点，因为我十五年前开始的，但你知道，可以说超过十年的安全数据，包括来自我们的MDR——托管检测响应服务，这里面有各种威胁和漏洞配对，还有攻击链之类的。
Well, maybe a little bit less because I started fifteen years ago, but you know, call it over a decade of security data, including from our MDR, our Managed Detection Response Service, that has all of this sort of threat and vulnerability pairings and sort of the attack chains.
我们基本上创建了海量的数据，并用来训练模型。
And we've basically created this massive amount of data that we use to train it.
所以我们在上面取得了很好的结果，而且我认为由于后期训练，它非常聚焦。
So we're getting great results on it and I think it is very focused because of the post training.
再说一遍，虽然它不能包治百病，但在安全方面能给你很好的结果。
You know, again, it's not going to cure cancer, but it's going to give you great results for security.
这是第一点。
That's number one.
第二点是安全和信任方面的问题。
Number two is there is a safety and trust aspect.
我们拥有所有客户的数据。
We have all of the data from our customers.
在过去十五年里，我们一直受到客户的信任。
We've been entrusted with that for the last fifteen years.
我们知道如何保护这些数据。
We know how to protect it.
我们知道他们想要什么。
We know what they want.
我们知道他们希望如何使用这些数据。
We know how they want to use it.
客户确实有这样一个担忧。
There's a real concern by customers.
他们不希望所有这些数据都流向实验室。
They don't want all this data going out to the labs.
他们不想把所有的敏感信息。
They don't want all their sensitive information.
他们不希望这样。
They don't want it.
不希望自己的信息被用来训练，对吧？
Their information being trained on, right?
所以，我们可以提供一种方式，让他们在自己的安全项目中获得所期望的信任和安全性。
So we can provide a way to get them the trust and security that they want for their own security program.
这种AI主权的概念对很多客户来说非常重要。
This sort of concept of AI sovereignty is very important to many of our customers.
所以我们并不是说排斥实验室，因为他们的技术非常棒，但你可以选择你想要的，或者可以在不同的模型之间进行路由，但最终我们想给客户选择的自由。
So again, we're not saying exclude the labs because they have incredible technology, but you can pick what you want or you can route between the various models, but ultimately we want to give customers choice.
我觉得你也提到了一点
I think you've also talked a little
关于政教分离的事情。
bit about separation of church and state.
你希望在一定程度上有一个中立的第三方，也能够，你懂的，去审查那些实验室提供的模型，或者市面上其他的开源模型，看起来CrowdStrike确实能扮演这个角色。
You want to make sure that to some extent there's a neutral third party that can also you know, interrogate some of the models that the labs are providing or some of the other open source out there, and you know it seems like CrowdStrike can definitely play that role as well.
我认为安全领域最重要的因素就是被信任，成为值得信赖的伙伴。
I think the number one factor in security is being trusted, being a trusted partner.
你可以看看市面上的那些供应商。
You know, you can look at the vendors out there.
显然有一些大公司在其中。
There's obviously a few big guys that are out there.
安全公司很多，但信任是点滴积累，却像水桶一样流失的。你知道，客户不想要那些可能把数据置于风险之中的冒险行为。
There's many security companies, but trust is built in drips and lost in buckets, and you know, customers, they don't want sort of cowboy activities that may put their data at risk.
而且，大家都知道，每个人都尽力了。
And, know, again, everybody's trying to do the best they can.
但我们整个重心，对吧？
But our whole focus, right?
而且我觉得在任何行业里，专注都特别重要。
And I think focus is incredibly important in any business.
我们的重心就是安全，就是保护客户的安全，我们不是要解决AI和其他领域的各种问题。
Our whole focus is security, and our whole focus is keeping our customers safe, and we're not trying to solve the world's problems in AI and other areas.
所以我觉得这绝对是我们的一个优势。
So I think that is certainly an advantage for us.
如果展望两年后，你觉得安全领域会是什么样子？
If you were to think out two years from now, what do you think is the state of the security world?
就是说从现在的角度看，会发生哪些变化？
Like what do you think shifted from this point forward?
两年之内，肯定会有很多本地模型直接跑在每个人的电脑上。
Certainly within two years you're going to have local models that are all running on people's computers.
你知道吗，可能混合使用本地模型，再加上最后一步接入前沿模型之类的。
Know, you may have a mixture of local models plus sort of last mile routing into into frontiers in terms of what they look like.
你会看到更多开放权重模型。
You're gonna have a lot more open weight models.
你只要看看过去六个月的代币变化就能明白这个趋势。
You know, you can just see the shift in the tokens, you know, over the last six months.
你看到的是什么变化，规模有多大？
What is the shift that you're seeing, and what's the order of magnitude there?
你觉得这种情况会持续多久？
How long do you think that continues for?
我几乎每天都要跟首席信息官们交流。
I talk to CIOs, you know, almost every day.
今晚我还跟纽约一家大银行的CIO吃了晚饭，他基本上说他们已经进入了AI领域。
I had dinner with one tonight from big bank in New York, and basically the comment is, you know, we got into AI.
他们全力投入，挺好的。
We went all in great.
但成本失控了。
The costs were out of control.
当大家收到高额账单后，狂用token就停了。
You know, token maxing, you know, stopped when people got big bills.
然后你猜怎么着，我说，那你们怎么看这件事？
And you know, I said, well, how do you think about this?
比如说，你们打算怎么办？
Like, what are you going to do?
他们说，这些模型真是…
And they said the models are so...
现在的开源权重模型真的好得不得了。
The open weight models right now are so good.
即使它们落后三个月，对我们来说也完全够用了。
Even if they're three months behind, it's plenty for what we need.
有个客户评论说，他们知道我对一级方程式赛车很感兴趣。
And one of the customers comments and they know I'm into Formula One.
他们说，银行需要的任何一级方程式相关活动，都会用前沿模型来处理。
They said any Formula One activity that we need for the bank will go to the frontier.
比如说如果是顶尖的模型，而且能赚钱，又特别重要，那我们就会付这个钱。
Like if it's Alpha and we're making money and it's like, you know, super important, we'll pay the money.
没什么大不了的。
Not no big deal.
但对于其他所有事情，比如真的很贵，我们会自己做所有的开源模型，并且对它们做更多的后训练。
But for all the other things like it's really expensive and we're going to create all of our own open weight models, you know, and do more post training on them.
他们就会这样运作。
Those sort of things that that's the way they're going to operate.
我坚信这一点，我看到一个又一个客户都是这样，他们都会走这条路。
I firmly believe that I'm seeing that in customer after customer that that's that's that's the way they're going to go.
我觉得当前环境中的一些挑战，部分原因是大家都看到了这一点，而且在顶尖模型层面，确实有一些担忧，不知道接下来会发生什么。
And I think, you know, part of some of the challenges in the current environment is I think people are are all seeing that and you know, there is some concern at the frontier level of what's going to take place.
所以我想关于开源模型，经常有人说它们至少部分是从那些实验室模型蒸馏出来的。
And so I guess on the open weight model, one of the statements that tends to be made is that they are at least in part distilled from the labs.
换句话说，中国的开源玩家实际上是在用Anthropic或OpenAI的模型，或者其他顶尖模型来训练他们的模型。
In other words, you have Chinese open source players who are effectively training their models against that anthropic or OpenAI model or other frontier models.
很多这些实验室并没有在尝试推出反蒸馏措施。
Many of those labs are not trying to roll out anti distillation efforts.
他们是在模型中构建防止被蒸馏的能力。
They're trying to build into the model the capability to not get distilled.
你觉得这会不会有点像网络攻防那样，能力不断升级，还是说你对两年内或者合理时间范围内的蒸馏状态有什么预测？
Do you think it's just sort of a back and forth where the capabilities keep going almost like offense and defense in cyber, or what's your prediction in terms of the state of distillation in two years or, you know, within some reasonable time frame?
我想先说的是，前沿模型在获取所有这些信息的时候，它们是在蒸馏什么呢？
I mean, I guess my first comment is what were the frontier models distilling when they got all that information?
我们都知道那些关于数据来源的诉讼，以及事情是怎么发展的，对吧？
I mean, we know all of the, you know, sort of lawsuits around where the data came from, and how that unfolded, right?
所以我们大概就是从那里开始。
So we kind of start from that.
这就像是精品酒一样。
So it's like a fine liquor.
它经过了三次蒸馏，你懂吧？
It's triple distilled, you know?
展开剩余字幕（还有 223 条）
确实是三重蒸馏。
It's triple distilled.
对，没错。
Exactly.
那你会问，蒸馏是从哪里开始的呢？
Like, so where does the distillation start from?
对吧？
You know?
所以我觉得事情就是这样，但实验室确实在努力防止蒸馏。
So I think it is what it is, but the labs certainly are focused on trying to prevent distillation.
显然，你看看市面上那些中国模型，大量蒸馏在发生，互联网上到处都有系统在四处抓取信息进行蒸馏。
Obviously, you know, when you look at the Chinese models that are out there, there's a lot of distillation takes there's factories of systems that are bouncing all over the Internet distilling this.
我觉得问题在于，秘密已经泄露了。
I think the challenge is that the genie is out of the bottle.
大量的蒸馏已经发生了。
I mean, there's a lot of distillations already taking place.
这些模型确实很好，但我们也要该表扬的地方表扬。
The models are very good, but we also have to give credit where it's due.
如果没有那么多芯片，就得想办法把资源用好，同时还要保持顶级模型的水平。
If you don't have all the chips and you have to be resourceful and you still want to have top models.
我觉得美国以外的地方也有一些创新，我们得注意这一点，对吧？
You know, I think there's some innovative things that are taking place outside of The US and we got to be mindful of that, right?
所以是的，这确实很难。
So yes, it's hard.
我觉得，想把蒸馏这个“精灵”重新装回瓶子里，会非常困难。
I think it's going to be hard to put the genie back in the bottle on distillation.
他们会尽最大努力，但你知道，很多模型现在已经……你觉得两年后会是什么样子？
They're going to do the best they can, but you know a lot of the models are are, you know, I think now you know what does it look like in two years?
我不知道。
I don't know.
现在你有很好的模型。
You've got great models today.
这些模型能用一阵子，里面已经打包了很多数据，所以我们就等着看怎么发展吧。
They can last for a while, and you have a lot of data that's already packed into them, so we'll see how it unfolds.
真希望我能预知未来。
I wish I had the crystal ball on that one.
哦，不会吧。
Oh, no.
只是有点好奇。
Was just curious.
然后我猜还有个……
And then I guess another...
预测这些总是很难，但我有点好奇。
It's always hard to predict these things, but I'm a little bit curious.
你觉得中国公司在开源或开放权重模型方面有没有结构性优势，能够保持在领先地位，还是说美国开放权重模型可能在短期内赶上？
Do you think there's structural advantages for Chinese companies to stay at the frontier of at least open source and open weights, or do you think US open weight models are likely to catch up in the near term?
嗯，确实有些开放权重模型要出来了。
Well, are some open weight models coming out.
我相信你，可能你也投资了这些公司。
I'm sure you you know, you may be invested in them as well.
我认为拥有优秀的美国开放权重模型将非常重要。
I think having great US open weight models is going to be very important.
所以，你知道，我们与英伟达合作的原因之一，当然是因为Neumotron是开放权重和开源的，不仅仅是开放权重，但拥有其他替代方案将非常重要。
So you know, one of the reasons why we worked with Nvidia was, of course, Neumotron is open weight and open source, not just open weight, but having other alternatives is gonna be very important.
而且我认为，再次，因为我在科技行业待得够久，你几乎会观察事情如何发展，就像互联网的操作系统是Linux，就这么定了，好吗？
And I think there's, you know, again, just being in technology long enough, you kind of look at how things unfold like the operating system of the Internet is Linux, like period, Okay?
这是有原因的，而且AI也会有一个操作系统，在我看来，它很可能是开源和开放权重的，到时候一切会如何发展。
And there's a reason for that, and there's going to be an operating system of AI and it's probably in my mind it's going to be, you know, an open source open weight when that happens and how it all unfolds.
我们拭目以待，但这并不意味着没有前沿模型的空间。
I mean, we'll see, and it doesn't mean there isn't room for any frontier model.
这也不意味着人们不会从中赚到很多钱。
It doesn't mean that people are not going to make a bunch of money off it.
你知道，围绕它的应用层也是。
You know the application stack sort of around it.
你知道那些工具链、系统怎么连接，以及怎么从企业角度思考这些问题。
You know the harnesses and how things are connected and how you think about enterprise.
这些东西都很重要，但最终来说，现在成本太高了。
All those sort of things are going to be really important, but ultimately like there's just too much cost involved right now.
你看很多前沿模型，它们其实都在被商品化。
When you look at a lot of these frontier models, I mean, they're being commoditized.
如果中国的模型落后三个月，很多公司就会觉得三个月不算什么……
If the Chinese models are three months behind, you know a lot of companies are going like three months is not...
我能从中得到什么？
What what am I gonna get out of it?
我……
I...
它能满足我所有的需求。
Does everything I want.
这就是挑战所在。
And that's the challenge.
最近，一家顶级金融公司，他们是个大型技术买家，告诉我，他们的组织根本没有能力每十天就吸收一个新模型发布。
A top five financial said to me recently, like a, you know, a large technology buyer there, that there was no chance that their organization had the capability to absorb a new model release every ten days anyway.
所以，落后三个月比他们组织接受的规范速度还要快，因为他们是受监管的。
And so three months behind, you know, was faster than the accepted well, like, regulated pace of their organization.
所以这多少对他们没造成什么影响。
So it made no difference to them to to to some degree.
我认为开源会继续存在，并成为解决方案的重要部分。
I I think open is here to stay as a big part of the puzzle.
我有一个问题。
You know, one one question.
我知道我们还会继续让你展望未来，这几乎是不可能的。
I I know we're gonna keep asking you to look into the future, and it's kind of impossible.
但考虑到你的安全背景，这是我在科技职业生涯中第一次听到一些非常聪明、受过良好教育的技术专家说，我们需要把互联网关掉。
But given your history in in security, this is the first time in my career in technology where some very smart, well educated technologists are like, we need to turn the Internet off.
我们只需要把所有东西都隔离起来，因为互联网明年就要崩溃，关键系统再连接它就不安全了。
We just gotta air gap everything because the Internet is gonna fall apart in the next year, and it will be unsafe for critical systems to touch it.
你觉得这种情况有多大的可能性？未来会不会有很多关键系统，社会都不敢让它们连接互联网？
How how much do you believe that to be, like, a reasonable probability in the future where there are lots of critical systems that society cannot rely on touching the Internet?
我觉得这纯属胡说。
I mean, I think it's nonsense.
你明白吗？
You know?
比如，安全问题，以及智能体可能做的事情，以及它们带来的风险，是不是真的那么严重？
Like, is security and and what agents can do and and the risk that they pose, is that serious?
当然。
Of course.
那是很严重的问题。
That's serious business.
我们都该正视这件事。
We should all look at that and go.
对，我们需要确保安全措施到位，但世界会因为这些东西而很快完蛋吗？
Yeah, we got to make sure that we lock things down, you know, but is the world going to end because of these agents anytime soon?
不会。
No.
而且你看这些智能体在做什么，我觉得大家对智能体有一种很神秘的理解。
And when you look at what the agents are doing, I think there's this sort of mythical view of what an agent does.
智能体不会直接从你的电脑飞到另一台电脑上。
Like an agent doesn't just beam itself from your computer to another computer.
它实际上就是在做人类会做的事情，只是做得更好、更快，对吧？
It actually just does the same thing a human would do, just better and faster, right?
所以你看看像 CrowdStrike 这样的公司长期以来在做什么，我们就是在防止这类攻击。
So if you think about what companies like CrowdStrike have done for a long time, we prevent against these sort of attacks.
现在这些攻击可能会来得更快。
Now they might come faster.
可能会有更多低级别的漏洞被串联起来，等等这些情况。
There might be more lower level sort of vulnerabilities chained together and those sort of things.
对攻击者来说是好事，对防御者来说不是好事，但我们正在构建能够应对所有这些攻击的技术。
Great for the adversary, not great for the defender, but it's building the technology that can deal with all these sort of attacks.
那么AI到底干了什么？
So what is AI done?
它让漏洞变得更多了，所以你会看到更多的漏洞，而且攻击来得更快，所以是更多、更快，但并没有出现那种全新的攻击类型，比如某个智能体突然就从你的电脑跑到另一台电脑上。我觉得大家有点想太多了，总以为会出现某种全新的技术手段。
It's created more vulnerability so you have more of them and you know you have more exploitation coming faster, so it's more and faster, but there isn't like a unique class that all of a sudden an agent just beams itself somewhere and I think people just get wrapped around the axle of like there's some new class of technique that's taking there.
其实并没有。
There isn't.
它只是在学习我们已经知道的东西。
It's just been trained on what we know.
它还没有自己发明新的攻击方式。所以在我看来，安全仍然是AI发展的推动力，这对安全行业来说非常重要，因为通常来说，安全是给业务踩刹车的。
It hasn't developed its own yet, And that's why I kind of look at it and go, security again is going to be an enabler for AI, which I think is incredibly important for the security industry, because normally it's, you know, you're putting the brakes on.
我们得花时间。
You know, we have to take time.
你不让业务跑得那么快，但现在大家都来找CrowdStrike和其他公司，问怎么让AI安全，这样我们就能在AI上走得更快。
You're not letting the business run as fast as they want, and now everyone's coming to CrowdStrike and others going, Well, how do you make AI secure so we can go faster in AI?
所以我是这样看的。
So that's the way I would look at it.
AI 已经是大势所趋了。
AI is here to stay.
我觉得有些担忧是存在的，也确实应该存在，但很多人对这些问题过于悲观了，我觉得这种担心未必有道理。
I think some of the concern is there, and it should be there, but sort of like there's a lot of kind of doomerism around some of this stuff, and I just don't think it's necessarily warranted.
我知道你有时也会以天使投资人的身份参与一些公司，还有生态系统的其他环节。
I know you're also, you know, sometimes involved with companies as an angel investor and, you know, other parts of the ecosystem.
除了安全领域，在AI领域里，还有哪些其他你非常看好的关键变化或重点方向吗？
Are there other areas that you're very excited about right now in terms of key areas of change or focus in the AI world in addition to the security world?
比如有哪些你觉得特别令人兴奋的事情，无论是刚刚发生的还是即将到来的？
Like what are some of the things that you think are really exciting that either just happened or are coming?
我觉得芯片以前不太时髦。
I think, you know, chips used to be kind of not in fashion.
当然，我们现在都知道芯片很热门，而且未来只会需要更多、更快、更专业的芯片。
Obviously, we know they are in fashion today, and you're only gonna need more of them and faster of them and more specialized.
而且我确实认为芯片领域会有一些重大突破。
And I do think there will be some big breakthroughs in the chip world.
说到速度，摩尔定律可能有点过时了，但归根结底，芯片出来后，东西确实变得更快更便宜了。
You know, when we think about just sort of the speed, take, you know, Moore's Law might be a bit data, dated, but at the end of the day, you know, the chips came out and things just got faster and cheaper.
我不知道具体的发展轨迹，但我可以保证未来也会是这样，越来越快，越来越便宜。
I don't know the trajectory, but I can guarantee you it's going to be faster and cheaper in the future like that.
这是显而易见的。
Can see.
所以随着这些专业芯片的开发，特别是在某些领域，随着围绕数学的新科学的发展。
So as these specialized chips are developed at, you know, focus in maybe particular areas as new science is developed around the mathematics.
要知道，性能可能会有阶梯式的提升。
Know there can be step function increases in performance.
功耗更低。
You know less power.
诸如此类的事情。
Those sort of things.
所以我觉得芯片是一个大领域。
So I think chips is a big one.
开源权重模型方面。
The open weight models.
我觉得还有很多增长空间。
Think I there's a lot of room to grow.
当然，美国模型也是，而且我确实觉得会有很多进展，能把这些模型缩小，然后比如说能在手机上运行。
Certainly The US models, and I do think there's gonna be a lot of advancements in taking these models and shrinking them down, you know, and and being able to run them on your phone as an example.
所以，我们大概都知道这段历史。
So I mean, we probably all know the history.
还记得以前发射第一枚火箭需要占一个足球场那么大的计算机房吗？
Remember the days where, you know, it took a football field sized room of computers to, you know, launch the space, you know, first rocket.
现在我们的iPhone里的算力比那整个房间里的还要强。
Like, we have more power in our iPhone today than they had in the entire room.
所以，这也会发生的。
So that's gonna come.
而且我觉得，如果你能让AI更快、更便宜、更高效，把它带到桌面上，带到手机上，在边缘设备上运行，这些都是很激动人心的机会，因为这才是真正的前景所在。
And I think the area where you can make AI faster, cheaper, more efficient, and you bring it to the desktop and you bring it to the phone where you're doing it at the edge, I think all of those are sort of exciting opportunities because that's really where I see.
你可能会有一个混合模型，但现在很多事情都可以本地完成。
You know, you're probably gonna have a hybrid model, but you can do a lot of it now locally.
再说，看看现在的TPU、MPU等等。
Again, just looking at the, you know, the TPUs, MPUs, etcetera.
就像，这些都是……
Like, these are...
它们都跑在我们今天已有的系统上，而且很多时候都被浪费了。
They're they're all on systems that we have today, and and a lot of times they're going to waste.
但我觉得它们被浪费的时间不会太长了。
But I I don't think they're gonna be going to waste that that much longer.
因为你每天的工作就是雄心勃勃地思考防守者能做什么。
Just because it is, you know, your job to think ambitiously about what the defenders can do all day.
我不会说，像AI防……
I I wouldn't say, like, AI def...
抱歉。
Sorry.
防守者...
Defenders...
安全防守者们在那种特定意义上算是悲观主义者，但他们总体上比较阴郁，对猫鼠游戏能否朝有利于他们的方向发展相当悲观。
Security defenders are doomers in that specific term, but they're, like, generally kind of dour and, like, pretty pessimistic about the cat and mouse game getting any better in their favor.
你能不能尽量论证一下，如果人们投资得当，我们做出更好的产品，并且我们自动化很多...
Can you can you make your best case for, like, actually, you know, if people invest correctly and we make better products and we automate many...
有没有可能真的在未来十年内，防守占据主导地位？
Like, is there a version where it's actually defense dominant over the next decade?
我认为可以，因为再说，就是那些并不一定新奇或独特的理念，但是能够始终迅速理解进攻方在做什么，并确保你的防守跟上，这将是关键。
I think it can be because if we again just with with the concepts that you know are not necessarily new or unique, but that the ability to always rapidly understand what the offense is doing and making sure that your defense is keeps up with it is gonna be the key.
这有点像很多年前我合著的那本《黑客暴露》书的现代版，就是说如果你向人们展示进攻是如何运作的，防守就会好得多。
I mean, it's sort of the the modern day version of the hacking exposed book that that I co authored many years ago, which was like if you show people how the offense works, the defense is gonna be a lot better.
那时候的节奏比现在慢得多，但我认为这将会很重要，而且作为行业，确保这是我们的关键任务，我在CrowdStrike团队中反复强调这一点，就是你必须给防守者提供与进攻方相同或更好的能力。
Now that was a much slower sort of tempo than it is today, but I think that's gonna be important, and as an industry, making sure this is a key thing for us, and I really drill this home with with our team at CrowdStrike, is you've got to give as good or better capabilities to the defender than what the offense has.
如果你只是稍微好一点，哪怕是数学上的优势，你仍然会占据优势。
And if you're just slightly better, you know, just mathematically, you're still gonna have an advantage.
而且我觉得，这始终会是一场猫鼠游戏，但关键是要玩得聪明。
And I think it's again, it's always gonna be a cat and mouse game, but being really smart about it is gonna be important.
今天要对抗AI，唯一的方法就是用更好的AI。
The only way to fight AI today is with better AI.
你知道，你得好好利用你手里的优势。
And you know, you've you've gotta use the advantages you have.
防御方最了解自己的系统。
The defender knows their own systems.
防御方手里有架构图。
The defender has the architectural diagrams.
防御方还有代码，对吧？
The defender has the code, right?
那你怎么把这些都利用起来，大幅提高自己比攻击方AI更聪明的概率呢？
So how do you put all of that into something that you can dramatically increase your probability of outsmarting the offensive AI?
我觉得唯一一个有意思的策略想法是，如果未来大部分代码都是智能体写的，那你其实可以借此推行更完善的安全架构。
The only tactical thing I I think is an interesting idea is the idea that you could, over time, if if it is going to be mostly agent written code, you could enforce better security architecture.
而且可能害怕漏洞，或者很多人把形式化验证一直看作是非常小众的东西。
And, you know, perhaps fear vulnerabilities or a lot of people where, you know, formal verification has always been enormously niche.
但即便不做到那一步，比如说，为什么所有软件都不安全？
But even even short of that, like, okay, why is all the software insecure?
一个很大的原因就是工程师数量有限，能力也有限，而且大家都在拼命赶速度。
Because, you know, one one big reason is that engineers are, you know, there's a limited number of them with limited capacity, and we're all trying to move very quickly.
但如果能以计算速度或愿意投入的资金来扩展，而且智能体有无限的耐心，也许我们最终真的能拥有更安全的软件。
But if you can scale at the speed of compute or your willingness to spend and agents have infinite patience, like maybe we could actually end up with more secure software over time.
这很有意思……
It's an interesting...
我觉得现在……
I I think there there is...
现在行业已经发展到这样的程度，代码质量可能真的足够好了，你肯定会得到一些好处。
Like the industry is at a point now where where the code is is probably good enough where you're gonna get definitely some benefits.
但同样，这些代码也在其他领域受过训练，比如它可能在有问题的数据上训练过。
But again, the code was was also trained in other areas like it was trained on stuff that you know may have had issues.
你知道吗，我自己也有开发者许可证。
So you know I have my own developer license.
我自己做一些应用，像Claude什么的，而且你知道吗，我有自己的安全代理来审查代码，它就相当于说，嗯，Claude刚刚创建了代码，但然后Claude现在用它的安全应用去检查，说那个代码不安全。
I build my own sort of apps and things like Claude and and you know, I have my own security agent that reviews the code and it's like, well, Claude just created the code, but then Claude is now with its security app going well, that code is insecure.
可是它刚刚创建的呀。
Well, you just created it.
为什么不一开始就这么做呢？
Like, why don't you just do that in the first part?
也许有一些优化空间，但我觉得人们不一定意识到的是，即使零漏洞，我希望这个行业能做到，你还是会有安全漏洞，因为现在很多人攻击的方式不一定是登录进去，而是利用身份问题。
You know, maybe there's some optimization, but I think what people don't necessarily realize is that even with with zero vulnerabilities, which I hope the industry can get there, you're still going to have breaches because a lot of what we see people today doing is they're not necessarily breaking in the logging in.
现在身份问题很严重，但归根结底，我们能把代码做得更好，减少漏洞，然后把这个好处传递下去，那就太好了。
There's such an issue around identity, but at the end of the day, we can get better code and we can have less vulnerabilities and we can ripple that through, that'll be great.
挑战在于，这是一个非常长的尾巴。
The challenge is it's such a long tail.
就拿微软举个例子。
You take Microsoft as an example.
就是因为所有这些。
Just because of all of the the.
他们想要保持的兼容性，向后兼容。
You know, compatibility that they want to maintain going backwards, backward compatibility.
有些东西一直没有被修复或现代化。
You know, there's things that never kind of get fixed or modernized.
你知道，这只是我们生态系统中的一个重要元素。
And you know, that's just one important element in our ecosystem.
但是想想外面所有其他的软件。
But think about all the other software that's out there.
你不可能一夜之间就把它们都修好，所以这会是一个很长的尾巴。
You're not going to just fix it all overnight, so it's going be a really long tail.
你已经运营CrowdStrike十五年了，在那之前，你是个企业家。
You've been running CrowdStrike for fifteen years, and before that, you were an entrepreneur.
安全...
Security...
我一直很喜欢安全领域，因为我们永远不会……
I've always loved security because, like, we're never...
作为生意人，这既是一个有趣又重要的使命。
As a business person, it's both an interesting and important mission.
作为生意人，这个问题不会消失。
As a business person, like, this problem is not going away.
但我感觉，很多大型成熟上市公司的高管现在都有点迷失方向，因为他们过去积累的优势正在受到挑战。
But I I feel many many CEOs of large established public companies are feeling a little bit unmoored in this period of time because so much of the value that they have built in their incumbency is is being challenged.
对。
Like Mhmm.
你怎么看待让自己保持与时俱进、让组织保持灵活，并且持续获胜？
How do you how do you, you know, think about keeping yourself up to date, keeping the organization agile, continuing to win?
对此有什么一般性建议吗？
Any general advice on that?
好的，我很乐意谈谈这个。
Yeah, happy to jump into that.
首先，我觉得你得认识到现在是一种不同的运营模式和节奏。
So first, I think you have to recognize it's a different operating model and cadence.
我们有成千上万的人，对吧？
You know, we've got thousands and thousands of people, right?
但你看那些刚起步的AI原生公司，他们只有几个人。
And when you think about some of the AI native companies that just started, they have a handful of people.
Claude全天都在运行。
You know, Claude is running all day.
他们让AI代理做各种事情，这就是新世界，挺好的，但你不能——不是说我们是那样——但你不能还抱着那种瀑布式的开发思维，事情要一年后才出来，那你就完蛋了。
They've got agents doing all kinds of stuff like that's the that's the new world, which is great, but you can't be in a not that we are, but you can't be in like the waterfall mentality of development, like things pop out a year later, you're going be out of business.
所以我觉得首先你得认识到，拥有庞大的团队和漫长的周期，那是不行的。
So I think you first have to realize that having these massive teams and these long schedules, like that is not going to cut it.
所以得有更小的团队，像小队那样，或者说是专注于某个问题的小组，他们可以利用AI，但你不能让几千人都来解一个问题。
So having smaller teams that are, you know, squads or you know, smaller groups that are focused on a problem, they can leverage AI, but you can't have like thousands of people show up to solve a problem.
你得有个更小的团队来做这件事，而我们已经做到了。
You got to have a smaller team to be able to do that, and we've done that.
我们在猎鹰卫士产品上就是这么做的，而且说到这点，我本人每周都参加三次会议。
We've done that with like our Falcon Guardian product, and to your point, like I was personally involved, you know, at three day a week meetings.
我还参与了它的设计。
I was involved in the design of it.
就像这样，我们在做什么。
It's like here's what we're doing.
改这个，改那个，你得回到创业初期的状态，因为需要那种程度的改变。
Like change this, change that, and you got to get back to your like startup roots because it takes that level of change.
你知道，在大公司里，有很多公司比我们大得多，而且你知道，你不能反应迟钝、慢慢悠悠的。
You know in a bigger company and there are many companies that are much bigger than we are, and you know you just can't be slow and plotting.
所以第一点就是要意识到运营模式和节奏在变化，你必须改变，必须在内部进行自我颠覆。
So that's number one is to realize that the operating model and cadence is changing and you got to change what you got to disrupt yourself internally.
然后你只需考虑，嗯，你的护城河是什么？
And then you just have to look at like, well, what are your moats?
如果你是记录系统，那护城河就相当深了。
If you're a system of record, it's a pretty big moat.
我们是安全领域的记录系统，对吧？
We're a security system of record, right?
而且我们拥有大家多年的数据。
And we have everyone's data for many years.
顺便说一句，我们是净数据创造者。
And by the way, we're net data creators.
我们的传感器，我们称之为安全代理，对吧？
Our sensors, we call them security agents, right?
但我的意思是，别把它和AI代理混淆了，就是那种终端代理软件，对吧？
But I mean, like not to be confused with AI agents, the software that endpoint agents, right?
那是一个净数据创造者。
That is a net data creator.
所以我们每天大概产生70亿个事件。
So we're creating 7,000,000,000,000 or so events like a day.
如果你不断创造这种新的数据流，那会给你带来优势，因为任何AI系统的命脉就是你喂给它的数据。
And if you're constantly creating this new data stream, that is going to give you an advantage because what's the lifeblood of any AI system is the data that you feed into it.
所以对我们来说，关键是想清楚我们的护城河是什么。
So I think, you know, for us, it's always like, well, what are our motes?
我们该从什么角度看？
How do we look at it?
我们怎么确保能颠覆那些老牌厂商？
How do we make sure that, you know we disrupted the incumbents?
我们颠覆了以前的格局，像麦咖啡那些老玩家，大多都消失了，或者被收购，或者转型了，你永远不希望自己成为被颠覆的那一方。
We disrupted the semantics, the McAfee's, you know, all those other players are like all gone for the most part or changed or bought or whatever, and you never want to be in a position where somebody, you know, disrupts you.
所以我们总在想，要不要自我颠覆？
So we're always like, do we disrupt ourselves?
我们怎么做才能与众不同？
How do we do things differently?
怎么才能换个思路？
How do we think differently?
但归根结底，我最关注的一点，也一直在跟团队强调，就是把客户照顾好，其他一切自然水到渠成。
But at the end of the day, you know, my main focus, and I tell this to our team all the time, if we take care of the customer, the rest takes care of itself.
客户想要的是得到保护。
What the customer wants is they want to be protected.
他们希望信任自己的供应商和合作伙伴，归根结底，希望做到非常经济实惠，没有人愿意花冤枉钱，所以用最合适的成本给他们最好的结果，并且体验上不需要七台控制台和六把椅子来回折腾，让它变得简单。
They want to have trust in their in their vendor and their partner, and at the end of the day, want to make it super economical like no one wants to spend unnecessary money, so give them the best outcome at the best cost and an experience that doesn't take, you know, seven consoles and six chairs to run around in and just make it easy.
这是最重要的一点，客户总是告诉我，部署你们的产品太容易了。
And that's the number one thing you know that our customers always tell me is like it is so easy to deploy your stuff.
今天早上我们刚开了个会。
Just we're at a meeting this morning.
会上有人就说，你知道吗？
It was like you know what?
有两家公司要合并。
We're there's two companies merging together.
他们决定用我们的方案。
They're gonna go with us.
另一家公司没用我们的，然后他们说，你们的东西部署起来太简单了，我觉得这一点很重要，就是实现价值的时间。
The other company didn't have us, and it was like, your stuff just is so easy to deploy, and I think that's a big one is time to value.
现在很多人都在讨论AI会导致失业，虽然这种事情实际上还没发生。
So a lot of people right now are talking about job losses due to AI, although that really hasn't happened.
我觉得到目前为止，反而净新增了就业机会。同样，也有很多人在问，我的孩子应该学什么，未来工作会怎么样，类似这样的问题。
I think the opposite has happened in terms of net job creation so far, And similarly, there's a lot of questions, what should my kids study, and what will happen in the future for jobs, and the like.
您对此怎么看，或者您会建议人们以后往哪个方向发展，或者这甚至是不是一个长期问题？
What what is your view on that, or where would you recommend people go going forward, or, you know, is this even a long term issue?
嗯，从人工智能的角度来看，如果你看看过去几百年发生的技术变革，每一次重大变化都创造了更多的机会和就业岗位。
Well, just in general from an AI perspective, and if you look the technology changes that have taken place over the last couple hundred years, like each major shift has created more opportunity and more jobs.
虽然岗位不同，但数量更多了。
They're different, but there were more of them.
我坚信这次也会是这样。
And I firmly believe that's gonna be the case here.
你看，不管是编程领域，你仍然需要人来管理这些智能体。
And, you know, when you look at whether it's programming, you're still gonna need people to manage the agents.
你仍然需要人来思考这些架构问题。
You're still gonna need people to think about these architectures.
这不是简单的一次性任务。
It's not just fire and forget.
我知道很多人喜欢这么说，但你还是需要人类的创造力。
I know a lot of people would like to say that, but there is human creativity that you're gonna need.
具体到网络安全，这是我经常跟年轻人、大学生们提到的一个领域，它确实是个很好的职业方向。
I think specific to cybersecurity, this is one of the areas that I always talk to young folks about, people in university, is it's a great field to be in.
如果你懂AI，又懂安全，这个领域不断在变，而且它其实没有终极解决方案。
If you understand AI, if you understand security, it's an ever changing field, and it's actually not a solvable problem.
我觉得每个厂商都希望能说他们解决了问题。
I think every vendor would like to say they solve the problem.
但实际上没法完全解决，你只能想办法降低风险。
It's not solvable, and you've got to mitigate the risk.
你需要想出最好的方案。
You've got to come up with the best solutions.
你要能保护好客户，这中间有很多因素要考虑，但事情总是在变化。
You've got to be able to protect the customers, and you know there's a lot of things that go into it, but it's always evolving.
未来五年会出现我们今天想都想不到的技术，如果你还在上大学，你们就会是思考如何保护这些技术的人。
There's going to be technology in five years that we haven't even thought about today, and if you're in university, you're to be some of the people that are going to be thinking about how you protect it.
所以我一直希望能回馈社会，也关心大学生们在做什么。
So I always, you know, try to give back and think about what college kids are doing.
我经常被家长问到，我的孩子该学什么？
I get asked all the time from parents, What my kids study?
我觉得网络安全、技术和人工智能是一个非常棒的领域。
I think cybersecurity and technology and AI is a fantastic field.
这是一个很好的结尾。
It's a great note to end on.
乔治，非常感谢你参加这次访谈。
Thanks so much for doing this, George.
谢谢参与
Thanks for joining
我们的节目。
us.
不用谢。
You're quite welcome.
非常感谢。
Thank you so much.
可以在推特上找到我们，账号是NoPriorsPod。
Find us on Twitter at NoPriorsPod.
如果想看我们的脸，就订阅我们的YouTube频道。
Subscribe to our YouTube channel if you want to see our faces.
在Apple播客、Spotify或任何你听播客的地方关注这个节目。
Follow the show on Apple Pod casts, Spotify, or wherever you listen.
这样你每周都能收到新的一集。
That way you get a new episode every week.
另外，还可以在no-buyers.com上注册订阅邮件，或者找到每集的文字稿。
And sign up for emails or find transcripts for every episode at no-buyers.com.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。