跳到主内容
@wquguru
精选82a16z(YouTube)产品与增长

Armanin:用AI红队进攻训练防御,发现90+零日漏洞

Building Cyber Defense for the Agentic Era

原文
发到 X
推荐理由

给出了具体的“红蓝对抗”落地路径:先用AI做红队挖掘漏洞(已验证发现90+零日),再据此构建防御,适合CTO或安全负责人参考其架构思路。

You don't have a defense unless you have a great offense to go up against. You want to be the Baltimore Ravens [music] defense of 2000. You kind of want to have your practice offense really push you. That's what Arman's going to do. We're going to be the all-star team on offense [music] coming at you so you can train your defense with what we're doing. You built mandate a great success. Why did you decide to get back on the field?

除非你拥有强大的进攻能力,否则你无法建立有效的防御。你希望成为2000年的巴尔的摩乌鸦队[音乐]防守组。你需要让训练中的进攻方真正对你施加压力。阿曼将做到这一点。我们将作为全明星进攻阵容[音乐]向你发起冲击,以便你能根据我们的表现来训练你的防守组。你打造的Mandate取得了巨大成功。你为什么决定重返一线?

I don't want to sit out the AI shift change when I've done 30 years in [music] security and the whole damn thing's about to change. What AI does in a microcond would take 70 humans. They can't even do it. It's apples to oranges. [music] This is a tsunami like has never been seen before in security.

我不想在这场AI变革中置身事外,毕竟我在[音乐]安全领域已经深耕了30年,而整个行业即将发生巨变。AI在微秒级时间内完成的工作需要70名人类才能做到。他们根本做不到。这完全是不可比的事物。[音乐]这场海啸般的冲击是安全领域前所未有的。

The whole let's slow down the models. We don't want cyber too late. The open models are already good enough. Armadin since January of this year, we have found over 90 zero days at customer sites all in production. This is not like rinky dink companies like these are like Fortune 500 companies. The differences or similarities between nation state attacks compared to AI today and then where you think AI can be in a couple years.

关于‘让我们放慢模型发展速度’的说法,我们不想让网络安全滞后。开源模型已经足够好了。自今年一月以来,Armadin已在客户现场发现了90多个零日漏洞,且全部处于生产环境中。这与那些小公司不同,这些客户都是财富500强企业。国家行为体攻击与当今AI之间的差异或相似之处,以及你认为几年后AI能发展到什么程度。

On [music] the defensive side, we're gonna say we're being attacked by these models, but we're not sure who's behind them. Is it a nation? Is it a human? Is it

在[音乐]防御方面,我们会说我们正受到这些模型的攻击,但我们不确定幕后是谁。是国家吗?是人类吗?还是它

Kevin, thanks for being here.

凯文,谢谢你来到这里。

No, thank you.

不,该谢谢你。

Okay, so you built Mandette.

好的,所以你打造了Mandette。

Yes,

是的,

obviously a great success. Um, many different chapters, you know, it ended up, you know, inside of Google ultimately. Um, why did you decide to get back on the field? It's a good question and you know I don't know if I decided it and that'll sound weird but I met with David Slater and with Travis Lam the other founders and Evan Pñena I knew uh you know I I could say they started this company they are the founders you know I met them they had the idea they pitched me on what they wanted to do and I saw the talent in them like Travis is a generational talent David Slater's and I mean this in a positive way freak nature, you know, like these guys are are really, really good.

显然取得了巨大成功。经历了多个阶段,你知道,最终它被整合进了Google。嗯,你为什么决定重返一线?这是个很好的问题,你知道吗,我不确定是否是我自己做出的决定,这么说可能有点奇怪,但我最近与David Slater、其他联合创始人Travis Lam以及Evan Pñena会面了。你知道,我可以这样说:他们创立了这家公司,他们是创始人。我见过他们,他们有这个想法,向我展示了他们的愿景,我看到了他们身上的才华。比如Travis是几代人难遇的天才,David Slater也是如此,我是正面意义上说的,他们具有非凡的天赋,你知道,这些人真的非常非常出色。

Evan Pñena is exceptional at what he does.

Evan Pñena在他所做的事情上出类拔萃。

And when you meet that team and you talk to them, the whole time I was listening to what they were doing, I was thinking, I want to be a part of this. You know, I don't want to sit out the AI shift change when I've done 30 years in security and the whole damn thing's about to change. You know, that's great. Everything I did is dead and everything else is new. Uh but meeting that team and they were starting the company and me realizing I'm I think I'm a real good fit with these guys um to accelerate the need like what Armadin is building every company needs now and I I was like this team that can build it and I think I can answer the now let's you know 30 years in security you meet a few people let's go to those people and say we've built what you need.

当你见到那个团队并与他们交谈时,在我倾听他们所做的一切的整个过程中,我一直在想,我想成为其中的一部分。你知道,我不想在这场 AI 变革中置身事外,毕竟我在安全领域已经干了 30 年,而整个该死的行业即将发生巨变。你知道,这太棒了。我过去做的一切都过时了,一切新的东西都出现了。嗯,但见到那个团队后,他们正在创办公司,而我意识到我觉得自己非常适合这些人,去加速满足像 Armadin 正在构建的那种每个公司现在都急需的需求。我当时就想,这个团队能把它建出来,而我认为我能回答‘现在’这个问题。你知道,在安全领域深耕 30 年后,你会遇到一些人,于是你去找到那些人,说我们已经构建了你们所需的东西。

Yeah. You know so I almost felt compelled to do it. I know that sounds weird, but I would not have founded a company again in my, you know, in mid50s and I was doing venture.

是的。你知道,我几乎感到有一种冲动要去这么做。我知道这听起来很奇怪,但在我的五十多岁时,在做风险投资期间,我不会再次创立一家公司。

It wasn't like, oh, I'm an entrepreneur and I love starting companies. That's not it.

并不是那种‘哦,我是个企业家,我喜欢创办公司’的情况。不是那样的。

And it wasn't, I'm not a VC. I'm just an operational guy. That's not it. I met the team and went, we have to do this. Yeah. I mean, that's really it.

也不是因为我是风投(VC)。我只是个运营人员。也不是那样。我见到了那个团队,心想我们必须这么做。是的。我的意思是,事实就是这样。

Yeah. And this whole AIC change was happening. Like that's the catalyst, right?

是的。而且这场 AI 变革正在发生。就像那是催化剂,对吧?

Yes. So yeah tell us about what Armanin does. So, Armadin leverages frontier models and AI on offense to test do you have exploitable risk and the re and that's what we do today. We call it armored in red. But we when we started a company Travis and and David Slater and Evan all knew the future of cyber security is going to be a the good guys have to build the offensive cyber cannon and shoot it at networks to make sure those networks can withstand these attacks because they're the ones that are coming.

是的。所以,是的,给我们讲讲 Armadin 是做什么的。Armadin 利用前沿模型和进攻性 AI 来测试是否存在可利用的风险,这就是我们今天所做的。我们称之为红队防护(Armed in Red)。但当我们开始创业时,Travis、David Slater 和 Evan 都知道网络安全未来的发展方向将是:好人必须构建进攻性网络武器,并将其射向网络,以确保这些网络能够抵御这些攻击,因为这些攻击正是来自那里。

But we also knew it's going to be AI on offense built by the good guys working and training with AI on defense built by the good guys. And you have to have both. So our act one was we got to be the best in the world at finding exploitable risk. If we're five minutes ahead of the bad actor, whether it be a nation state, criminal nuisance, if we're five minutes ahead of them, we also recognized our act too, armed it in blue.

但我们也知道,未来将是好人构建的进攻性 AI 与好人构建并协同训练防御性 AI 相结合。你必须两者兼备。所以我们的第一幕是,我们必须成为全球在发现可利用风险方面最顶尖的团队。如果我们比不良行为者——无论是国家行为体还是犯罪骚扰者——领先五分钟,只要我们领先他们五分钟,我们也认识到我们的第二幕是蓝队防护(Armed in Blue)。

We got to stop it. compensating control tourniquet.

我们必须阻止它。补偿控制止血带。

And so that's what Armadin does. We're building the force field you will need in the AI age to defend yourself from AI attacks.

这就是 Armadin 所做的。我们正在构建你在 AI 时代所需的力场,以保护自己免受 AI 攻击。

Yeah. Tell [snorts and clears throat] us about the nature of the capabilities of AI attacks today and then where you think it goes.

是的。告诉我们 [鼻息声并清嗓子] 当今 AI 攻击能力的性质,以及你认为它将走向何方。

Well, great. So the nature of them is first off, we're getting a weird window in time where we're seeing them, but not at the same level you'd expect. Like I've seen nothing like what Armadin's already built in the wild, which is there's 25,000 agents all in concert, all working together doing really, really smart things without going on bizarre fishing trips. Because when you respond to an AI attack, you can tell it's AI very quickly.

好吧,太棒了。首先,它们的性质在于,我们正处于一个奇怪的时间窗口,能看到它们,但并非以你预期的那种水平。比如,我在野外从未见过像 Armadin 已经构建的东西那样——有 25,000 个智能体协同工作,一起做着非常、非常聪明的事情,而不会去进行那些奇怪的钓鱼行动。因为当你应对 AI 攻击时,你可以很快识别出那是 AI。

At least I can because I've done, you know, I've thought about a lot of offense. I've responded to a lot of attacks in the past that were led by humans,

至少我可以,因为我做过很多进攻方面的思考。过去我应对过许多由人类主导的攻击,

right? and a human goes to point A, then to point B, then to point C through their intrusion. AI does little things like four or five differences, but one would be it'll break into point A, then laterally move to point B, next thing you know, it's trying to break into point A again.

对吧?人类在入侵过程中会依次到达 A 点、B 点、C 点。而 AI 则做一些细微的事情,比如存在四五个差异,但其中一点是:它会先攻破 A 点,然后横向移动到 B 点,紧接着它又试图再次攻破 A 点。

Yeah.

是的。

It's like, I get the drone swarm, but you could probably coordinate and think a little bit better. And that's that's where it's at today. It'll get better and cleaner. Um, but the differences are first and foremost, uh, the scale of what AI can do dwarfs humans. like in ways humans don't even get right.

就像我说的是无人机蜂群,但你本可以协调得更好、思考得更周全一些。这就是目前的情况。它会变得更好、更精炼。嗯,但首要的区别在于,AI 所能做到的规模远远超过人类,甚至在人类都搞不对的方式上也是如此。

Um so you have a scaling problem in that humans could always find only one path into a network.

嗯,所以这里存在一个扩展性问题,因为人类始终只能找到进入网络的唯一路径。

Yeah. They had to be selective because they had to devote their limited resources to one direct path. Right.

是的。他们必须有所选择,因为他们必须将有限的资源投入到一条直接路径上。没错。

Yes. And and then uh so scale is a challenge. Speed ridiculous. What AI does in a microscond would take 70 humans they can't even do it. It's apples to oranges. And then so what was always lacking is is AI creative

是的。然后,规模是一个挑战。速度更是荒谬。AI 在一微秒内能完成的事,需要 70 个人类去做,而他们甚至做不到。这完全是不可比的两回事。而且,一直以来缺乏的是 AI 的创造力。

or effective? But when it comes to what we do, u we don't need the fanciest model. We're not trying to speak 400 languages, you know, with our models and all that kind of thing. What Armenan's doing on offense is we're finding vulnerabilities, exploitable risk. That is code. That's a structured language, a structured process. Because it's structured, AI is going to be great at it. Right.

还是有效性?但在我们做的事情方面,我们不需要最花哨的模型。我们并不试图让我们的模型说 400 种语言之类的东西。Armenan 在进攻方面所做的是:寻找漏洞、可利用的风险。那就是代码。那是一种结构化的语言、结构化的流程。正因为它是结构化的,AI 在这方面将会非常出色。对吧。

Right. So, I really think it's already here today. Like the whole let's slow down the models. We don't want cyber risk too late. the open models are already good enough and and these things are common now. It's just a matter of the minute you have anonymous availability of GPUs, you'll see far more criminal attacks.

没错。所以,我真的认为它今天就已经存在了。比如整个‘让我们放慢模型发展’的说法,我们不想让网络安全风险来得太晚。开源模型已经足够好了,这些东西现在很常见。问题只是,一旦匿名可用的 GPU 出现,你就会看到更多的网络犯罪攻击。

Oh, interesting.

哦,有趣。

Yeah. You know what I mean? But until you can attack anonymously and you know, it's hard to do crime when people know your name. It's better to if you can if you can commit a crime anonymously, here's my tip for criminals. If you can commit a crime anonymously, that's a lot smarter than doing it with your jersey on with your name on it. And uh so anyway the diff the difference in attacks and what we're seeing now we are at the precipice first inning still of AIE attacks coming um and I think that's just because of the cost and availability of the models is not as readily available to the crim criminal element as it will be in the future.

是的。你明白我的意思吧?但在你能匿名发动攻击之前,当人们知道你的名字时,犯罪是很难进行的。如果可以匿名犯罪,这是我的建议给罪犯的:如果你能匿名犯罪,那比穿着印有自己名字的球衣去犯罪要聪明得多。总之,目前我们看到的攻击与我们即将面临的 AIE(人工智能增强)攻击之间的差异在于,我们仍处于第一局早期阶段,我认为这仅仅是因为模型的成本和可用性尚未像未来那样容易为犯罪分子所获取。

Yes. Exactly.

是的。没错。

Okay. So you your experience in working in the security industry for 30 years,

好的。所以你在安全行业工作了30年的经验,

you um probably saw a fair amount of nation state [snorts] attacks, right?

你可能见过相当多的国家支持[喷鼻息声]的攻击,对吧?

Every day.

每天都有。

Every day. So talk about the differences or similarities between nation state attacks and I use that just to say the

每天都有。那么谈谈国家支持攻击之间的差异或相似之处,我使用这个术语只是为了表示

原文超出正文长度上限,此处截断——上游还有内容,完整版见上方「原文 ↗」。

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

关联信息,但可能不是同一事件