EP278 The Agentic SOC: Are We Measuring Time Saved or Risk Reduced?

你好。

Hi there.

欢迎收听谷歌云安全播客。

Welcome to the Cloud Security Podcast by Google.

感谢你今天加入我们。

Thanks for joining us today.

我们来自云未来大会2026现场。

We're coming to you from Cloud Next twenty twenty six.

我们现在在普华永道的展台。

We're in the PwC space.

今天的主持人是我，蒂姆·皮科克，还有安东·切巴克。

Your hosts here are myself, Tim Peacock, and Anton Chewbackend.

今天我们也请到了普华永道英国公司的高级顾问马特·格雷格森。

We're joined today by Matt Gregson, a principal at PwC at UK.

是这样吗？

Is that right?

你是在英国吗？

Your UK place?

在美国。

US.

在美国。

US.

哦，对，不过你可以剪掉我的口音。

Oh, yes, but you can cut off my accent.

我可以留着它。

I can use it.

这个挺酷的。

Cool around that one.

好的。

Alright.

安东，第一个问题你来问吧，我觉得。

Anton, you have the first question, I think.

我有。

I do.

接下来就是我的。

And it's next.

还有一些关于智能防御和智能SOC的公告。

There are also some announcements about agentic defense, agentic SOC.

我想先提个很简单的问题。

And I kinda wanna start like a very softball question.

在你看来，目前智能SOC的前沿水平如何？你见过哪些案例？

In your opinion, what's the state of the art of agentic SOC or what have you seen?

我觉得蒂姆也想问，你见过最智能的SOC是什么样的？

And I think Tim wanted to ask, what's the most agentic SOC you've seen?

你可能不用回答这个问题，但我觉得这么问挺合适的。

So don't answer that one maybe, but I feel like it's a good way to phrase it.

对。

Yeah.

我觉得今天公布的这些消息都很令人振奋。

So I think some of the announcements we've seen today, they're exciting.

我们正逐渐看到，这种智能安全运营中心真的要成为现实了。

We are gonna get to a I think we're starting to see how this agentic SOC is gonna become a reality.

目前我看到AI的应用，不只是智能代理，大概分两种方式。

I think right now, well, I'm seeing sort of two ways of how agentic's being used and and AI in general rather than just agentic.

一种是AI作为协作者，分析师可以通过聊天界面问问题，比如我的检测结果有哪些、告警情况如何，还能生成趋势报告。

I think we've got a component of a AI as a copilot type of scenario where analysts can ask through a chat interface questions about what are my detections, what are my alerts, things like that, get some trend reports.

这种功能是通过Gemini平台实现的，它和SecOps平台的MCP服务器对接在一起。

And that's being enabled through, like, a Gemini platform where it's tied to an MCP server of of of a SecOps platform.

这是我目前看到的主要试点模式。

So that's the core pilot type model I'm seeing.

另一些方面，很多组织还在试点阶段，或者刚开始考虑怎么把智能代理接入生产环境。

And then on the second side, I'd say this is still in a lot of organizations in, like, pilot mode or or just starting to think about how we get into production.

关键问题是，现有的SOAR流程和工作流里，该在哪儿插入智能代理的部分？

And that's more about where did my SOAR quite pipelines and workflows, where do I plug in the agentic piece?

对。

Now Mhmm.

对某些客户来说，可能是由一个大语言模型在经过我的确定性流程处理后，结合威胁情报、主机和身份信息做出决策。

For some clients, that might be a an LLM making a decision after being passed through my deterministic pipelines, the context of threat intel and the the host and the identity.

我觉得在其他情况下，可能更需要思考：我们该怎么实施隔离？

I think if there's other cases, it may well be trying to think through, okay, How do we do containment?

怎么让智能代理采取行动？

How do we get the agent to take action?

我觉得很多组织还在摸索，怎样才能建立足够的信任，来全面放开这种我坚信我们所有人都期待并认为必要的完全自主方案。

And I think organizations are still trying to feel through how do we get comfort before we we allow that fully autonomous vision that I think we're all looking forward to and and thinking we need

朝这个方向迈进。

to get towards.

我认为包含自动修复的完全自动化还很罕见。

I think full auto including remediation is rare.

我也会这么认为。

I I I would be like Yeah.

很少见。

Rare.

对。

Yeah.

就像

Like

你今天是在争取‘最低调陈述奖’吗？

Are you vying for understatement award today?

我这是在极力 understatement（低调表达）

I'm vying for the understatement

今年

of the

最低调的说法，Bryce。

year, Bryce.

这次，我无论如何都要拿下它。

This time, I wanna get it after all.

但整个流程是从自动处理到确认，再自动转交给人工团队来解决。

But it is the full auto to confirmation, full auto to hand off to a team of humans to resolve.

我的意思是，就算在最棘手的时候，有时候也完全是自动化流程。

I mean even in the sore days there was full auto workflow sometimes.

我记得以前当分析师的时候，遇到钓鱼邮件，系统有时会自动删除邮件。

I remember again from my analyst days phishing phishing email response would sometimes go full auto to the email deletion.

所以，在安全领域的一些特定场景里，全自动流程是行得通的。

So full auto has worked in security in certain campaign boxes.

没错。

Yes.

所以我们只需要扩大这些系统的处理范围，让环境更适应全自动操作。

So we just make the boxes bigger and make the environments more amenable to full

这就是我所说的‘代理型行为的可接受范围’。

This is what I call the agentic Overton window.

这是指哪些代理行为是被允许和接受的界限。

This is the window of possible acceptable agentic response behavior.

我认为，作为整个行业，作为开发智能代理系统的软件供应商，我的目标是在未来三年内扩大智能代理的可接受行为范围。

And I think our mission as an industry and as a software vendor who builds agentic things, my mission is to increase the size of the agentic Overton window over the next three years.

所以我想聊聊智能代理的优势。

So I want to talk about agentic and the upsides of it.

我本人并不算特别聪明。

And I'm not very smart.

所以我喜欢问一个很简单的问题。

So I like to ask a very simple question.

当我们变得智能化之后，什么会变得更好？

What gets better when we become agentic?

无论是对SOC团队、SOC负责人、CISO，还是董事会——无论你关注哪个层级，在智能型SOC中，哪些方面会得到改善？

For a SOC team, for a SOC leader, for a CISO, for a board, whichever of those four levels you want to talk about, what gets better in an agentic SOC?

我觉得这要看你怎么看待这个问题。如果是董事会的角度，我认为使用智能型SOC能降低风险。

I think depending on which way you're viewing this, if you're the board, I think your risk is going to get reduced by the use of an agentic SOC.

我觉得响应时间和修复时间都会显著缩短，如果我们能做到这一点

I think your time to response, time to remediate, it's going to really increase and if we can do that in a

针筒上的凹痕会自己弹回去。

way The dent would shrink out of syringe.

应该会缩回去，对的。

Should shrink, yeah.

不好意思。

Sorry.

这里的好处在于。

The goodness here.

信息是开放的。

Message is open.

好处会增加，但这才是我们真正关心的，对吧？

The goodness would increase but The is what we're looking for, right?

我觉得我们就能达到这个目标。

And I think that's where we can get to.

随着模型运作方式的不断演进，以及威胁环境的变化，

I think as we see a lot of evolutions in how the models are working and the changes in the threat landscape.

这意味着我们可能会看到更多零日漏洞，

That means we're going to see maybe more zero days possibly, which

确实有可能。

means Possibly.

当然了。

I mean, certainly.

没错。

Certainly.

对的。

Yeah.

这意味着客户会遭遇安全事件。

That means clients going to have incidents.

我们会遇到具备AI能力的攻击者，能在环境中迅速横向移动、提升权限。

We're going to have threat actors who are AI enabled that can move through an environment very, very quickly, lateral movement, privilege escalation.

所以我们需要引入智能代理来替我们自动执行修复操作。

So we are going to need that agentic piece to fit in to take remediation actions on our behalf.

为了快速澄清一下，在别人指责我们制造炒作之前，我想说，最近有些人对‘AI驱动的攻击者’这个说法特别敏感，但背后其实有非常真实、可衡量的现实：如果有人有个零日漏洞工厂，转动手柄就能产出零日漏洞，明天就要一个，再转一下就出来了——这确实是AI驱动的攻击者，跟那种得花六个月研究才能搞出一个硬核零日漏洞的人，完全不是一回事。

As a very quick point so before people blame us for proliferating hype I feel like occasionally people get triggered newly triggered by the whole AI enabled attacker but there is very hard measurable reality behind what we meant If somebody has a little zero day factory where they crank the handle of a model and zero day comes out and they need one tomorrow and they crank the handle and it comes out, it is very much an AI enabled attacker and it's very different from the person who needs to research for six months to get a really hardcore just zero

得花上二十五万美元。

spend a quarter million dollars.

或者直接花二十五万美元买一个。

Or just spend a quarter million dollars to buy one.

如果那些‘神话代币’定价合理，他们可能真会去那个零日工厂里转一圈，买一个回来。

Which they might do turning the franc in that zero day factory depending on how those mythos tokens are priced.

没错。

True.

所以这次，我们真的在讨论一个实实在在的问题。

So in this case we are genuinely we are talking about something very genuine.

这可不是有人开玩笑说的借口。

This is not an excuse as somebody joked.

哦，那是个高级攻击者。

Oh, it was a sophisticated attacker.

现在我们可以说，哦，这是个由AI驱动的攻击者。

Now we can say, Oh, it was an AI enabled attacker.

那我们能说什么呢？对。

What can we Yes.

这其实已经在发生了。

That's actually already happening.

我见过，但我要说的是，这有明确的事实依据，就是下游层面已经发生了变化。

Have seen But from my this is something very precise, grounded in fact that is the changes in downstream stuff.

是的。

Yeah.

但很多时候，当我们谈论AI驱动的攻击者时，他们用的还是那些广为人知、有文档记录的攻击路径，对吧？

But in a lot of cases, though, when we talk about AI enabled attackers, they're still using kind of like well known and documented like attack paths, right?

他们还是在 dumped LSATs，而我们已经有办法检测这些了。

They are still dumping LSATs and we have ways of detecting that.

所以我觉得，现在大家有点过于悲观了。

So I think we have there's a lot of doom right now.

对吧？

Right?

我们几乎因为AI攻击的走向而陷入恐慌，但我认为还是有希望的，因为一旦攻击者完成初始入侵，我们仍有能力发现这些攻击路径。

You know, we're in meltdown almost over some of the where this is going, but I think there is some hope in the sense that we have the opportunity to detect these paths once the initial access has been obtained

是的。

Yeah.

我们可以利用这一点，我认为那个智能安全运营中心的功能会真正发挥作用。

And we can use that and that agentic SOC piece I think will come through for us.

我差点就说，有积极的恐慌和消极的恐慌，当然这话听着别扭，我真正想说的是，现在确实存在过度炒作，但背后也有令人担忧的现实，比如我们刚才提到的补丁更新滞后问题，所有这些现实都让我思考：人类在未来到底该扮演什么角色？有时候我会想象，我们已经身处2030年的安全运营中心，或者说是2030年。

I was this close to saying yes there is good doom and bad doom and of course that line does not land so what I really want to say is there is scary hype but there's also scary reality behind some of this, like we talked about the recent event about the patching speed being behind and but all the reality of it leads me to think what would be the role of humans and as I sometimes say imagine that we are in 2030 SOC or 2030

侦探角色，而这距离现在只有四年。

Detective Which is only four years away,

顺便说一句，团队明智地提醒过，这并不是什么遥不可及的未来，好像我们要飞去火星一样。

let's Which is by the way team wisely reminded that it's not some kind of remote future where we fly to Mars.

不，这离我们很近。

No, it's quite close.

在这种情况下，人类到底做什么呢？

What do humans do in that

我能插一句吗？

Can I give you a brief aside?

我有时候会这样。

I sometimes do this.

我特别喜欢词语。

I love words.

‘末日’这个词原本并不只是指坏事。

The word doom didn't originally just mean bad things.

它只是指结果。

It just meant outcome.

所以‘末日’以前其实是……

And so doom actually used to be

我本来可以

I could

已经发生了

have done

好的结局这件事。

the good doom thing.

八百年前，你本来可以有好的结局。

You could have good doom eight hundred years ago.

所以给我们讲讲，什么是好的结局

So please tell us about the good doom

和坏的结局

and the bad

四年后的人类结局，不是八百年前的。

doom four years from now, not eight hundred years ago.

那么人类到底在做什么？

So what do the human do?

人类的结局会是什么？

What's the doom of the human?

我觉得人类还是会挺好的，毕竟人类还存在，这挺幸运的。

I think the human is gonna be well, we still have humans, thankfully.

对，这一点我清楚。

Right, I'm aware of that.

但我认为，现在看到的很多营销宣传，一级角色正在消失。

But I think there's a lot of some of the marketing we see right now is tier one's going away.

我觉得一级角色会被重新定义，包括我们如何使用这些分析师以及他们的运作方式。

I think tier one is gonna be reimagined how we're gonna use these analysts and how that operates.

我觉得这更像是一种代理监督者的角色，当代理采取行动来隔离主机或控制身份时，需要有人来 oversight。

I think it's more of an agent supervisor type role when agents have taken action to maybe contain hosts and contain identities.

不过，我们还是希望有人来复核这些操作。

Well, we still want a human to review that.

我们需要检查代理都执行了哪些操作。

We wanna review what actions agents have taken.

代理的行动速度非常快，所以还是需要一个代理监督者的角色。

They are taking actions at great speed, So there's a sort of an agent supervisor role.

我觉得在检测工程和自动化方面，还有更大的发挥空间，得把这些体系搭建起来。

I think there's more of a role around detection engineering and automation, trying to build that out.

这不再是单纯的分诊角色，而是更偏向监督、自动化工程，还要紧跟威胁态势的变化，确保我们设计的检测规则和自动化流程能跟上威胁的演进。

So it's less of a triage type role, more of the supervision, automation engineering, and really staying close to the threat landscape as well and how that evolves so that we are building those detections and those automations and making sure we've got the agents able to respond to that evolution of the threat landscape.

我待会儿要做一件特别奇怪的事。

I'm going have something really odd to do.

我们这边一共三个人。

So we are three of us.

我们在这个领域已经干了挺长时间了。

We've dealt with this domain for quite some time.

有些人头发都花白了，但心里头，咱们还是2030年的状态。

Some are you know, have gray hairs and all that, but deep in our hearts, it's 2030.

那如果人类手动做告警分诊，是不是就不用穿袜子了？

Does it mean that there are no socks when humans do allert triage manually?

不是。

No.

不可能。

No way.

不好意思问这个，我真的不知道该怎么回答。

I'm sorry I'm asking this because I honestly don't know what I would answer.

真的吗？

Really?

但你觉得未来几年它们都会适应吗？

But You think they will all adapt in the next

人们还是会用 PHP 维护大型机。

People will four mainframes in PHP.

没错。

Exactly.

对啊。

Yeah.

所以我才觉得，四年之后还是会有真人负责一线告警处理。

That's why I think there will still be humans doing tier one in four years.

是的

Yeah.

我觉得十年后还是会有人在做一线工作

I think there will be humans Like doing tier one in 10

我们刚才在聊末日，现在我真有点沮丧了

we were talking about Doomsday and now I'm actually depressed.

末日还没来，但这一幕真让我受不了

Doomsday did not do it, but this bit did.

我觉得我们都希望在这件事上能来场变革，但恐怕它只会慢慢演变

I think we all want a revolution in how we do this, but it's probably gonna evolve.

对

Yeah.

对吧

Right?

事情就会这么发展

That's how it's gonna do.

而且现在还有好多机构连基于硬件的多因素认证都没用上，漏洞补丁也打不全。

And we still have organizations out there who still don't have fish resistant MFA, still not patching.

你看，行业里还有很多问题没解决，别以为点一下按钮就能搞出一个全能型的智能安全运营中心，把所有问题都搞定。

Like, we have a lot of problems in the industry that we've not tackled, and the idea that we're just gonna click a button and then we've got an agentic SOC that solves all that.

我们还得先做好那些基础的安全工作，才能真正享受到智能安全运营中心带来的好处。

We still have to do the hard security work to get some of the benefits of the agentic SOC.

安东，你老说IT世界里最强大的力量是什么？

Anton, what do you always say is the strongest force in the IT universe?

IT占你股份的比例

IT in

你的份额。

your share.

对。

Yes.

我在Gartner干了八年，学到的最重要一课就是：IT在你公司所占的股份比例，是宇宙中最强大的力量。

That was my main lesson after eight years at Gartner, is that IT in your share is the strongest force in the universe.

所以四年之后。

So four years from now.

但我听现在这拨人说，因为有了AI智能模型，这种力量非常强大，虽然我们有点跑题了，但问题确实存在。

But I'm listening to this time it's a different crowd that says because we have AI intelligence models, there is a pretty strong and we're going way too far afield but it's We're way kind off questions.

这次是另一拨人，声音特别大、特别响，还不只是在推特上。

This time it's a different crowd, it's very strong and very loud and not only on Twitter.

我听到了他们的声音，其中一些人有计算机科学博士学位，还有一些是AI科学家，我不能轻易忽视他们，所以我不断对自己的‘IT惯性’观点产生怀疑。

So I hear them and some of them have PhDs in computer science, some of them are AI scientists and I can't just dismiss them so I constantly have second thoughts about my point about IT inertia.

说实话，我原本希望你能说，到2030年，没人再手动处理告警垃圾了。

So I honestly hoped you would say that by 2030 nobody does alert trash by hand.

抱歉，我刚才有点走神了，我本来以为到二月就会这样了。

I'm sorry I'm channeling my not me internally because I kind of thought by February

你这话说得可真少见地乐观啊。

It's unusually optimistic for you.

确实特别乐观，你说得对。

It is really unusually optimistic, you're right.

我不太明白这到底意味着什么。

I wonder what it means.

我不知道昨晚飞去拉斯维加斯的是哪个安东，但肯定不是我认识的那个。

I don't know which Anton flew to Vegas last night but it's not the Anton I usually know.

我想问个问题，关于帮客户升级他们的袜子。

I want to ask a question about helping clients modernize their socks.

去年我们做过一期特别棒的节目，采访了在银行工作的马尼莎，她和普华永道一起推进了这个现代化项目。

We did an amazing episode last year with Manisha at a bank where she worked with PwC for that modernization.

你每天都在帮人做这些事。

You're helping people do this day in and day out.

你怎么判断一个人是不是准备好用智能代理了？

How do you know when somebody's ready for agents?

是不是得达到一定成熟度，才能玩转智能代理？

What's the like, you must be this mature to ride the agent ride?

我觉得要充分发挥智能型SOC的作用，需要好几个要素协同配合。

I think there's a few different components that come together to get the most out of an agentic type SOC.

你可以从零开始，也能在智能代理上取得一些成果。

And you can start from nothing and you will have some success at agents.

但对我而言，如果有一些前提条件，这些成果还能更大，对吧？

But for me, that success could be greater if you had some prerequisites, right?

这样智能型SOC也会更持久。

More sustainable of agentic SOC as well.

首先，得有一个数据平台，把主机和身份的相关数据都集中起来，这样才能做必要的信息补充。

I think first of all, it's having a data platform where we have the data in there about our hosts, our identities so that we can do the enrichment that we need.

我们还需要把核心服务整合到一个平台里，以便能快速对防火墙和身份提供商做出响应。

I think we need our core services integrated into a platform that allows us to very quickly take actions in firewalls, take actions in IDPs.

对了，你刚才说的是‘平台’。

By the way, you said platform.

你没说‘多个平台’或者‘一堆工具’。

You didn't say platforms or many tools.

这一点说得真好。

This is a really good point.

所以你是在强调集中化，而不是去中心化。

So you hint at centralization rather than decentralization.

我退一步说。

I'll pull back.

对我来说，当我们把智能代理用在SOC里时，我觉得仍然很有价值的是：先由告警触发预设的响应流程，自动执行确定性操作，完成数据丰富和威胁情报查询这些基础工作，然后再交由智能代理处理。

And for me, like, if we think about the when we apply the agent in the SOC, I still see value in alerts being triggered that executes a playbook that takes deterministic actions, does the enrichment, does the threat intelligence lookups, all that good stuff, then it engages the agent piece.

如果你有一套成熟的问题调查流程，就能更好地引导代理，给它提供清晰的指令。

And if you've got good processes about how you investigate that you can sort of prompt the agents on and give them that,

我觉得这才是成功的关键。

I think that's the key to success.

所以，在你看来，一套良好部署和落地的SOAR系统，是AI驱动的SOC能否成功的衡量标准？

So having good, properly deployed and implemented SOAR in your mind is a success criteria for AI SOC?

这确实是个值得深入讨论的话题。

This is probably something that's worth discussing.

我并不是说你错了，我其实同意你的观点，只是担心有人一听就激动起来，拿着火把和叉子冲过来。

I don't want to say no, you're wrong, I agree with you, but I'm afraid people would show up with pitchforks and say

对。

Yeah.

确实有人觉得，一旦警报触发，就可以启动代理，它就能正常运作，对吧？

And there's certainly a view that, you know, once an alert triggers, you can execute an agent and it will work, right?

钱。

Money.

令牌，抱歉。

The token Sorry.

按容量计费的令牌。

Token By capacity.

我们使用

We use

我们用‘令牌’这个词，其实说白了就是钱。

this euphemistic term, tokens, but we really mean money.

货币令牌，各位听众。

Monetary tokens, listeners.

这背后的成本是多少？再想想，假如那个代理没有按你预期的方式运行，就得让人工介入。

The cost of that, and then if you think about, like, let's say that agent does not execute the way you want to, and then your human needs to come back in.

到时候，人工还得从警报触发的那一步重新开始。

Your human is gonna start back from the beginning at the alert.

你其实希望人工能快速介入检查，如果之前已经完成了信息增强，所有内容都集中在一个地方——也就是那个统一的视图，我觉得这依然很有价值。

You actually want the human to be able to very quickly go in and inspect, and if you've already got that enrichment, that it's executed and all in one place, that single pane of glass, there's still value in having that, I think.

你们俩在午餐时聊了一个特别有意思的观点，就是关于速度的问题。

So you two made a really interesting point in your conversation over lunch about speed.

能不能跟我们的听众分享一下，你们对AI速度和确定性SOAR操作速度的看法？

Could you share with our audience your perspective on AI speed versus deterministic SOAR operations speed.

你们当时具体是怎么论证的？

What was the argument you made there?

我们当时在讨论，如果假设未来零日漏洞会越来越多，导致客户遭遇入侵，攻击者一旦获得初始访问权限，接下来肯定还要提权、横向移动，这些步骤一个都少不了。

Well, we were talking about how if we work on the basis that we're going have more zero days, that will lead to incidents at clients and that they gain initial access, there will still need to be privilege escalation, lateral movement, all those activities.

这些步骤很可能都会借助AI来完成。

Those are gonna be AI assisted likely.

这意味着我们的反应速度会非常快。

That means we're gonna move very, very quickly.

这些行为和手法中，有些会产生大量告警，但我们能通过确定性的自动化流程迅速响应，很多时候，这种方式比依赖智能代理还要更快。

Now some of those behaviors and techniques are gonna be noisy that we can respond to very, very quickly using deterministic pipelines, and I think that's gonna be a quicker way of operating in a lot of cases than applying the agents.

我觉得智能代理确实有其用武之地，但关键是要搞清楚在哪些环节真正该用它们。

I think agents have absolutely got a role to play, but we need to evaluate where do we apply the agent in those techniques.

我觉得这里有个很有趣的地方，听众们，人类反应最慢，智能代理快一些，但确定性的流式检测或响应流程可以快到极致，还不受令牌数量的限制。

I think the interesting thing in here, listeners, is there's some kind of hierarchy of humans are very slow, agents are quick, but deterministic streaming detection pipelines or deterministic streaming response pipelines can be very, very, very fast and not constrained by tokens.

这正是我们在Narasay展会上用的那张图：AI模块放在中间，而确定性自动化才是我们理想中该迁移过去的部分，哪怕一开始是从AI入手。当时现场很多人都很激动，连连说不行不行，我们就是要让AI做最终决策。

And this is literally the slide we have at the Narasay show was that AI box was kind of in the middle and deterministic automation was something that you ideally want to move some stuff to even if you start with AI and all the people were very triggered, said no, no, no, we want AI to be the ultimate.

但归根结底，如果你用AI来生成确定性自动化流程，而且它真的有效——这一点很重要——那它会更便宜、更快、更好。

But ultimately if you use AI to create deterministic automation and it works, this is important, it is cheaper, faster and better.

安东，你的意思是说AI并不是万能的，它只是个工具，关键是我们得聪明地使用它？

Are you, Anton, implying that AI is not the end all be all and it's another tool that we should be intelligent on how we use?

这想法可真够颠覆的。

That's a radical thought.

不，这个想法确实挺激进的，所以我现在得对这种激进的想法谨慎一点。

No, that's a radical thought, so I'll be a little careful with the radical thoughts at this point.

但没那么激进的一点是，有些东西是要花钱的，或者像你提到的，会消耗令牌。

But the less radical part of this is that some of the stuff costs money or as you pointed out, tokens.

令牌。

Tokens.

但你描述的这种架构，是把大语言模型放在流水线的后端，我认识一些人——也不算太疯狂，就是特别热情乐观，他们想把模型往更前面推，直接对接原始数据。

But you described an architecture where an LLM is in the chain towards the later end of the pipeline, I met people, maybe not quite crazy people, but very enthusiastic, optimistic people who want to shift it left towards raw data.

我们之前请过一位嘉宾，

We had a guest

就是这么做的。

who was doing that.

上一季我们请过一位嘉宾，把云审计日志直接接入他们的大语言模型。

We had a guest last season who was piping cloud audit events into their LLM.

你还记得那位嘉宾吗？

Do you remember that guest?

我记得，他基本上平安无事，还把这事讲了出来。

I do and mostly he survived to tell the tale.

是的。

Yeah.

不过你该问下我们现在的嘉宾下一个问题了，我们已经跑题太远了。

So you should ask our present guest next question though because we've gotten very far afield.

刚才问的是什么来着？

What was the question about?

就是为什么要把大模型放在流程后端，以及成本和操作的依据。

The logic behind LLM being later in the chain, the proof of cost and activities.

你想啊，当警报触发时，我们那些流程里，有些可能只是自动关闭或者自动修复。

If you think about when an alert triggers and we have those pipelines, some of those pipelines may just lead to auto close or auto remediate.

这种情况下根本不需要大模型。

We don't need an LLM

确实不用。

for that.

不需要动脑？

No brain needed?

我们画个流程图吧？

We flowchart?

我们没必要把AI用在每一件事上。

We don't have to apply AI to absolutely everything we do.

我们有办法检测某些类型的威胁，而且对这些情况很有把握，可以直接采取行动。

We have ways of detecting certain types of threats and we have high confidence in those that we can take that action immediately

对。

Yes.

至于

Regard

这些东西已经奏效了，那些也能奏效，而且确实有效

stuff this stuff has worked, and this stuff can work, and it does have

得继续运行，不能停摆。

to be to do didn't stop working.

谢谢。

Thank you.

没错。

Exactly.

对。

Yeah.

并不是所有人都已经做到了这一点。

And not everybody has got that in place.

并不是所有人都建立了能自动执行这些操作的SOAR剧本。

Not everybody has built out these SOAR playbooks that take those actions.

对。

Right.

还有很多组织还处在安全成熟度的较低水平。

There are people still living below that line of security maturity capability.

有些组织遇到警报后，分析师还得手动去查威胁情报，一个个登录不同平台去查找信息。

Work with organizations where they have alerts and then analysts jumps into the alert and then starts doing the manual threat intel lookups, doing all sorts of different portals, looking things up.

这其实是个绝佳的机会，可以先搭建这样的工作流程，再想想在哪里引入智能代理，把人从重复操作中解放出来，转而承担监督和把控的角色？

Like, that's a great opportunity to build out that sort of workflow first and then figure out where can I apply the agent in there to take the human out of the loop and more on top and looking at the sort of the supervisory role?

有道理，不过我想再转个有点特别的话题。

That makes sense, and I wanna shift in a somewhat odd direction again.

确实是这样。

It is Again.

虽然这个问题是我提前准备好的，但今天在飞机上我读了一本关于AI的书，作者在书里抱怨人们总是把AI当人来看，还提到大家普遍对此有顾虑，可他突然话锋一转，说其实我在公司里更愿意把AI当成同事——我知道我刚才一边说别把AI当人，一边又这么想，但事实是，当我真把AI当同事看待时，工作效率反而明显提升了。

It is from the prepared list, but oddly enough, on the flight today, I guess, I was reading a book about AI, and in the book, the author was complaining about how people anthropomorphize AI and the usual brought the usual hesitation, and then he suddenly said, but actually when I work in my company whatever I prefer to think of AI as a teammate and I know I just rambled about not anthropomorphizing but if I think I noticed that practically if I think of AI as a teammate I just get better results.

这让我特别有共鸣，也让我想到了SOC环境里的AI同事。

And this to me was really interesting and it brought up in a SOC AI teammate.

如果你在实验室里有个智能代理、机器或者现代AI系统，你会不会把它当成一个同事，然后直接问它：‘你为什么这么做？’

Like if you have an agent, machine, robot, something in the lab or modern AI brain, would you think of this person as kind of a teammate and you can say, hey, why did you do this?

你为什么阻止了这次访问？

Why did you block access?

或者，你会不会把它纯粹看作一个机器任务、一个机器人？——等等，我刚才又在把AI当人了，抱歉，我是说，你会把它当成一个工具吗？

Or would you think of him, of it, sorry, now I'm anthropomorphizing, would you think of it as kind of a machine workload or robot?

这有关系吗？

And does it matter?

哦，这有关系吗？

Oh, does it matter?

没关系这一点很重要。

It doesn't matter is important.

对。

Yes.

我喜欢‘队友’这个概念，就是那种代理型的AI，这其实跟我之前说的呼应上了：在SOC里，你确实有这种像队友一样的AI，同时在流程和工作流里也有其他AI。

Because I do like the concept of a a teammate who is sort of an agent, and I think that goes back to what I said earlier that what I'm seeing is you have that teammate type of AI in the SOC, and then you've got the AI within the the pipelines and the workflows.

如果你能有一个AI代理，几乎像副驾驶一样，能接入你的威胁情报平台，了解你的安全运营体系，我觉得这样提问、获取答案就特别自然，就像队友一样，嘿，你能……

And I think if you can have that AI agent that is almost like a copilot with you that is clued up into your threat intelligence platforms, it is clued up into your security operations stack, I think that is a good way of asking questions, getting answers, and almost like a teammate, Hey, can

帮我看一下

you check

这个吗？

this, please?

你帮忙看一下，好的。

Can you check this, yes.

今天要处理的任务是

Tasks today

那些占时间又得交出去的事，我觉得这才是真正靠谱的职责，而且现在完全能做到。

that, you know, take up a lot of time and you've gotta delegate them, like, that I think is a legitimate role and something that is achievable right now.

对。

Yes.

至于一些工作流方面的功能，我觉得我们还得再往前走一步，得先搞清楚治理机制，明确我们到底希望它做什么，而不是只靠那个协作者功能。

Whereas some of the more of the workflow things, I feel like we need to get a little bit further down the path on, like, better governance and and better sort of, like, understanding and and and criteria of what we want it to do as opposed to the copilot piece.

我喜欢这个想法。

I like it.

说个有趣的事，周一我在客户顾问委员会问了一群安全负责人：你们怎么准备迎接这些效率超高但又特别天真的新队友加入你们团队？

A funny aside, I asked a group of security leaders at our customer advisory board on Monday, how are you preparing for these incredibly productive, yet very naive teammates to join your security teams?

我一个同事马上就说：蒂姆，别担心这个。

And one of my colleagues immediately says, Tim, don't worry about it.

有你在，我们没问题。

We do fine with you.

要特别感谢温迪，开了个很好的玩笑，拿我逗乐了。

So that shout out to Wendy for a very good joke at my expense.

我想问个关于幻觉的问题。

I want to ask a question about hallucinations.

作为行业，我们更该担心AI幻觉，还是未来团队成员的疲劳问题？

Should we be, as an industry, more worried about AI hallucinations or human fatigue in our socks of tomorrow?

我觉得情况会更糟。

Think this is gonna be worse.

上下文还是不够扎实，不是说哲学层面的普遍问题。

Context is still soft, not philosophically in general.

对，不是普遍情况。

Yeah, not in general.

就处理告警这个问题来说，我们更该担心幻觉，还是疲劳？

Literally for problem of responding to alerts, are we more worried about hallucination or fatigue?

我觉得问题更出在别的地方，也许‘幻觉’这个词用得不太准。

I think it's more in the and maybe hallucination is the wrong term.

在一些新模型里，我看到的幻觉现象比以前少多了。

I'm not seeing as much hallucination in some of the newer models as we previously were.

我更多看到的是在数据特别杂乱的环境里， telemetry 信息太多，AI 就理解错了方向。

What I am seeing is more in very noisy environments where there's a lot of telemetry, the AI gets the wrong end of the stick.

这可能是最好的应对方式。

Probably best way.

某些行为在那个客户的环境里其实是正常的，但 AI 却判定它一定是恶意的。

It seems certain activity, and that is normal in that customer's environment, whereas and it's concluded it must be malicious.

它缺乏更全面的上下文来真正理解情况。

It doesn't have the broader context to understand actually

我上次查的时候，这叫‘过度响应’。

That's called the most positive last time I checked.

对，不过

Well, yeah, but

从智能代理的角度来看，我们总觉得随便丢什么给它，它都能神奇地给出答案。

From an agentic piece, we seem to feel like we can just throw anything at it and it'll just magically come up with

说得对。

So correct

我觉得这才是更值得担心的地方。

I feel like that's sort of more the concern.

如果我们想想现在把大语言模型用在哪些地方，目标是让它们更自主，那基于这种误报或者其他错误，它们会做出什么决定呢？

And if you think about where we're applying the LMs, if the goal is to have more autonomy with them, what decisions will be made based on that sort of false positive or whatever it is?

对我而言，问题更在于这种‘幻觉’类型，而不是人类的疲劳。

So for me, feel it's more of the hallucinations style versus the human fatigue.

还有一个相关的问题是，人类对AI的疲劳感现在进展到什么程度了？

And one follow-up to that is where are we on humans getting fatigued of AI?

目前情况怎么样？

How's that going so far?

这是Google Next大会，大家都挺开心的。

It's Google Next, we're all happy

现在我们居然不能谈AI疲劳了，可我们明明在说这个。

to see We're not allowed to talk about AI fatigue Now we're

可以谈了，但我觉得还没到那一步，你还没到那儿。

allowed but I don't think we are quite there You're not there yet.

说实话，我不是想替你说话，但说实话，有些地方我们才刚摸到边儿。

Honest, I'm not trying to put words in your mouth, but I I mean, we're scratching the surface in some cases.

这个行业每个小时都在发生新的变化。

There seems to be evolutions in what's happening in the industry every hour.

这真的让人特别兴奋。

I it's really exciting.

我觉得我们这次真有机会解决一些长期存在的重大安全问题。

I think we've got a real good shot at solving some real big security problems that we've had.

对。

Mhmm.

过去几年我们经历了无数次大规模数据泄露，每次都说这次不一样了。

You know, we've gone through very, very large breaches over the last few years, and we always say this time it's different.

我们真得赶紧把这事抓起来。

We've really gotta get on top of this.

但我觉得，面对这些新兴模型，我们真得落实那些早就该做的纵深防御策略了。

But I feel like with some of these emerging models, we really do have to execute on some of the defense in-depth plans that we've had.

人工智能在其中将发挥巨大作用。

AI is going to play a huge role in that.

但这并不是唯一的作用。

It's not the only role.

我们还得把那些一直被提及的基础安全措施真正执行到位。

Like, we have to execute on some of the fundamental security practices that we've been talking about for

已经很久了。

a long time.

想想海瑟那期节目，对吧？还有谷歌的海瑟·阿特金斯，她对这些工具和机器有非常务实的看法：在某些特定领域，我们确实能产出安全的代码。

To remember the Heather episode, right, and I think her other public speaker Heather Atkins from Google she has a very well grounded opinion that with these machines, with these tools, we can produce only secure code in certain buckets.

再说一遍，我不是那种空谈的人，别太信我，但海瑟是。

Again, I'm not an abstract person, don't trust me, but Heather is.

如果我们能取得哪怕一部分相同的效果，我觉得所谓的AI疲劳根本就不是问题。

Feel like the AI fatigue is not a thing if we achieve even some of the same results.

够了。

Enough.

也许今天市场上确实有正面的需求，所以我不觉得

Maybe you've got some kind of positive market demand on today, so I don't

知道，是吧。

know Yeah.

真正该担心的是你家里的供水安全。

Really worried about what's in your water supply.

对。

Yeah.

我们该开始考虑关闭它了，但我认为它的价值依然存在。

We should start thinking about closing it, but I think that the value argument is still there.

你之前问过什么在变好，但时间节省总是关键——典型客户都希望省时间，或者他们的团队太忙，或者人手不够。

What gets better, you already asked, but the time savings come up as typical clients who want to save time or maybe their team is overworked or doesn't do enough.