Common Defense：一則會議邀請，怎麼變成駭客入口？

0:00先別信訊息：AI 時代的社交工程

0:00哈囉，哈囉。早安，早安。今天大家怎麼樣？很好。來談 AI，好的跟壞的。好，我是 Alex，來自 Common Defense，Quantstamp 的資安部門。今天我要說服你們為什麼不該相信任何訊息，以及在 AI 時代怎麼偵測、防禦詐騙或社交工程。嘿，好久不見。Hello, hello. Oh, good morning. Good morning. How is everyone today? Good. Well, time to talk about AI, good and bad. So, yeah. All right, well, I'm Alex from Common Defense, cybersecurity arm of Quantstamp, and today I'm going to convince you why you should trust no message and how to detect and defend against scams or social engineering attempts during the age of AI. Hey, it's been a while.

0:52有多少人在 Telegram 上突然收到這種訊息，還可能是幾乎沒講過話的人？對，我看到有人有同感。如果你收到這種訊息，你大概會驚訝裡面直接放了 Calendly 連結，邀請你去預約通話。如果你天真，真的會中招，去把 Calendly 約好。邀請會進你的 email 跟行事曆，然後你可能還會回對方一句好聽的，那個正在對你做社交工程的人：「喔好，很高興聽到你的消息。How many of you received a message like this on Telegram, out of the blue, from somebody you barely spoke to potentially, right? So... Yeah, I see some people can relate to that. Yeah, and if you receive a message like this, well, you'll be surprised that, you've probably been surprised that there is a Calendly link right there, so inviting you to actually join and book a call. And if you're naive, you would actually fall for that, and you will book the Calendly invite. It will land on your email and calendar, and then you would maybe write something nice to that person trying to social engineer you, saying, "Oh yes, great, great to hear from you.

1:48待會見」之類的。然後你進 Zoom，覺得有點怪。可能是畫面有點不對勁，或聲音不太對。為什麼？因為這是駭客。他想讓你裝東西，然後 Telegram 上又來一則：「嘿，Zoom 品質不好，改用 Microsoft Teams。」連結寫著去下載 Teams。你打開連結，可以看到這是假連結，因為 Teams 被故意拼錯。Yes, talk soon," and so on. Then you join the Zoom meeting and you see something odd, right? It could be either the video being a little weird or the audio doesn't sound quite right. Why is that? Well, because this is a hacker. He's trying to get you to install something, and you get a message like this on Telegram, saying, "Hey, Zoom quality is bad. Let's switch to Microsoft Teams." And there is a link there that says, you know, you download the Teams. You open the link, and as you can see, there is a fake link because Teams is misspelled intentionally.

2:37你下載了，就被駭了。這種事可以發生，而且過去已經發生很多次。為什麼？很單純：Web3 安全不再只是把智慧合約顧好。Vitalik 剛才也說，需要安全的金融。對，我們有設計上安全的區塊鏈，但你的筆電不是。光 2025 年，從個人筆電被駭走加密資產的案件就增加了，數字超過 15.8 萬，這是很大的數字。And you download that, and you are hacked. That's what can happen, and that's what happened many times in the past already. Well, why is that? It's simply because Web3 security is no longer just about securing smart contracts, right? So, as Vitalik was saying, there is a need for secure finance. And yes, we have blockchains that are secure by design, but your laptop is not. So in 2025 alone, there has been an increase in hacks of crypto from personal laptops, and the number exceeded 158,000, and that's a huge number.

3:27作業安全損失已超過 DeFi

3:27想像有多少人只因為資產在某台筆電上、或沒顧好，就丟了資金。2025 年假冒詐騙增加了 14 倍，作業安全失敗造成的損失超過 20 億美元。這是很大的數字，因為是 DeFi 的四倍。這跟 2021 年很不一樣，那時我們很多人在台上講的是 DeFi 被駭、橋被駭。最近這幾年，情況不太一樣。為什麼現在待在這個領域這麼難？Just imagine how many people lost their funds just because it was on someone's laptop or maybe just not properly secured. Impersonation scams increased 14 times in 2025, and more than $2 billion were lost due to operational security failures. And that's a huge number because it's four times more than DeFi, which is a different picture from 2021, when a lot of us were on stage and saying that, you know, displaying the issues of DeFi hacks, of bridges and so on. And nowadays, in the last few years, the situation is a little bit different. And why is it so difficult to be in this space at the moment?

4:18因為 AI 帶來很多新的複雜度。就在幾天前，有一則很紅的推文，有人說接到一通假裝是他太太的電話，太太說要錢去加油，但她開的是 Tesla，所以顯然不是真的。可是聲音聽起來一模一樣，如果你不夠警覺，非常容易中招。假身分證件。也許老一輩跟人講話時，會比較習慣請對方出示證件，一拿到證件就安心說：「好，你看起來是真人。」Well, there has been a lot of new complexity due to AI. Just a few days ago, there has been a viral tweet by somebody who claims that he got a call from somebody pretending to be his wife, and the wife needed money for filling up the gas tank, but she drives a Tesla, so it was clearly not the real message. But nonetheless, the voice sounded exactly the same, and it's very, very easy to fall for something like that if you are not alert enough. Fake personal IDs. Maybe some older generations could be more comfortable when they speak to somebody and ask for the ID from that person, and once they get an ID, they might be comfortable and say, "Oh, okay, you seem like a real person."

5:18現在不行了。現在要偽造任何證件都很容易。你只要用一個沒有那種安全機制的 LLM。這也幫不上忙。這正在變成大問題。而且你可以便宜、大規模地發動攻擊。以前如果有人要私訊這個房間裡的每一個人，大概五則、十則之後就會被停權，因為內容都一樣、同一個 IP。現在假設這個房間有 200 人，就會有 200 則個人化訊息，用的是每個人公開可查的背景。But not anymore, right? You can fake any ID very easily nowadays. You just have to use an LLM that doesn't have that type of security feature. And yeah, so this doesn't help either. Now it's becoming a big issue. And you can launch cheap attacks at scale, right? If previously someone would have to message everybody in this room, you'll probably get banned after five or ten messages, right? Because all of them would look the same. It would be from the same IP address and so on. But nowadays, if we have, let's say, 200 people in this room, there will be 200 personalized messages that use everyone's background that is publicly available.

6:03這些還可以從不同地點發出。這就是 AI 驅動攻擊能規模化的原因。如果有人經歷過早期網路，2000 年代初的社交工程大概是一封假裝奈及利亞王子的 email，你大概還有時間慢慢回。你只是讀一封信，沒有急著要回。現在不一樣。你們剛看到的這次嘗試，駭客會帶你跨不同 App。從 Telegram 開始，你上鉤了，會議邀請就進 email。And all of that can also be launched from different locations, right? And that's what enables AI-driven attacks at scale. And if some of you are from the early phases of the internet, where the social engineering in the early 2000s would look something like an email from someone pretending to be a Nigerian prince, and you could probably, you probably have enough time to respond to that, right? You just read an email, and there is no urgency for you to respond to that, right? But nowadays it's different. As you've seen in this attempt, hackers move you across different apps, right? It starts with Telegram, then if you're hooked, the meeting invite arrives on your email.

6:55跨 App 切換與有組織駭客

6:55然後也會進你的行事曆。Zoom 通話期間，又會在 Telegram、Zoom，也許還有別的 App 之間來回切。這些會讓你放下戒心。這種程度的情境切換，你跟不上。現在頻道跟應用程式就是這麼多。另一個例子：你在市集買了東西，送貨的人請你提供 OTP 才能收貨，但那個 OTP 其實是銀行寄來的。很多人就是這樣中招。Then it would be on your calendar as well. And then during the Zoom call, there will be a back-and-forth switch between Telegram and Zoom and maybe some other app. And all of that just makes you lower your guard, right? You just cannot keep up with this level of context switch. And there are just so many channels and applications nowadays. And another example of that is hackers. If you bought something on a marketplace and there is a delivery person asking for OTP for you to receive the delivery, but what can happen is that OTP is actually from your bank. Right, and that's how many people fall for that, right?

7:42你其實預期 Amazon 送貨會有 OTP，手機上又真的看到某個 OTP，不夠警覺的話，就可能把錯誤的 OTP 講給錯誤的人。另一層複雜是：駭客行動變得有組織。不再是地下室某個失業的人坐著隨機釣魚。這是企業，是商業運作，要把它看成一家大公司，每個人有自己的角色。某處有資料仲介，在找你公開暴露的資料。You just, you actually expect an OTP for your Amazon delivery, and you do see some OTP somewhere on your phone, and if you are not alert enough, you might also say the wrong OTP to the wrong person. And another complexity is that hacks become organized. It's no longer somebody unemployed in the basement sitting trying to just phish for some random benefit, right? It's an enterprise. It's a business operation, so it needs to be seen as a big company where everyone has their own roles, right? There's a data broker somewhere who is looking for the publicly exposed data about you.

8:30然後會有人研究怎麼對你做社交工程。會有人生成要寄給你的 AI 內容，會有英文毫無破綻的來電者跟你講話，讓你不起疑。最後如果成功，他們會從你身上抽走一些錢。同事推薦過一篇有意思的文章，是 ANY.RUN 的沙盒實驗，證實其中一些駭客行動的組織性質。我很推薦去讀，很有洞察。Then there will be someone who researches how to actually social-engineer you. There will be someone who generates AI content to send it to you, and there will be a caller with flawless English talking to you, so that you don't suspect anything. And eventually, if they are successful, they will extract some money from you. And an interesting article was suggested by my colleague. It's about a sandbox experiment by ANY.RUN, which confirms the organized nature of some of these hacks. I highly recommend reading this article. It's very insightful.

9:16社交工程有哪些階段？大概可以拆成三個。第一次接觸，是他們在 Telegram 找上你，幾乎像一則陌生開發。你一接話，下一階段就開始，你跟這個所謂的駭客會有更多來回。執行階段就是你真的拿到惡意連結，真正的攻擊發生在這裡。那我們能做什麼？哪個階段能真的擋住傷害？第一次接觸？對，最理想。What are the phases of social engineering? Well, maybe we can kind of split them up into three phases. The first contact, that's when they reach out to you over Telegram, almost like a cold reach. And if you engage with that, that's when the next phase starts, and we can probably see a lot more interaction back and forth between you and the alleged hacker. And the execution phase is exactly when you actually get the malicious link, right? That's when the actual attack happens. But what can we do about that? At which stage can we actually prevent the damage? In the first contact? Yes, ideally.

10:04在第一封訊息就擋下來

10:04如果根本看不到第一則訊息，就沒有互動，也就完全沒有社交工程的空間。接話階段呢？還是有機會擋住傷害。好處是互動會帶來更多脈絡，因為駭客至少會有十幾則訊息，你有更多上下文，駭客在不同應用上的足跡也可能更多。執行階段。那可能太晚了。還是不會？下一張再看。If somebody does not see the first message in the first place, then there is just no interaction, so there is no potential for social engineering at all. During the engagement phase? Yes, it's still possible to prevent the harm. And the good thing about that is that there is a lot more context from the interaction, because there will be at least a dozen messages from the hacker, and you have a lot more context, and maybe the hacker has a lot more footprint across different applications from your interaction. The execution phase. That might be too late. Or is it? We'll find out in the next slides.

10:50解法是什麼？第一個：如果能把通訊頻道統一起來，會很好。中間可以有某種偵測防火牆，坐在被保護的使用者跟所有可能的通訊頻道之間。Telegram、WhatsApp，也許連 Zoom 跟 email。想像你所有通訊頻道都這樣被保護。今天稍早講了很多隱私，這個解法如果能自架，而不是把這件事交給第三方，也會很好。So what are the solutions? Well, the first one is it would be really nice if we manage to unify communication channels, right? There might be some sort of detection firewall, which would sit between the protected user and all the possible communication channels, right? So it will be Telegram, WhatsApp, maybe even Zoom and email. And yeah, so pretty much imagine having all your communication channels protected in this way. And since we spoke a lot about privacy just earlier today, it would be also nice if that solution would be self-hosted, instead of trusting a third party with that.

11:41怎麼實作？很天真的做法是：為什麼不把我在社群媒體收到的每一則訊息，都丟進公開 LLM？它分析訊息是不是詐騙、有沒有可疑，能力很好。有用嗎？小規模也許有用。一個人也許有用，一小群人也許有用。但一到幾百個使用者，可能撐不住。你得付很多錢買 AI token 才能這樣做。就算付了，公開 LLM 也可能把你限流。And how do we implement that? Well, the very naive approach is: why don't we send every single message that I receive on my social media through a public LLM? It has very good capability to analyze a message for scam, right, or any suspicion. Does it work? Well, it might work for small scale. Maybe it works for one person. Maybe it works for a small group of people. But once you go to hundreds of users, that might not be sustainable, right? You will have to pay a lot of money for AI tokens in order to be able to do that. And even then, you might be throttled by public LLMs if you do it this way.

12:28所以這條路不通。當概念驗證也許可以，長期不行。那怎麼辦？我們在前面加一些規則。某些訊息，尤其是陌生開發，只要指定 pattern 就能偵測。例如我稍早那則：「嘿，好久不見。」如果那則訊息用雜湊存起來，你就能當 pattern 偵測，甚至在請求進公開 LLM 之前就做決定。跟 OTP 有關的訊息，也可以標成需要小心。這有用嗎？好一點，但還是不太能規模化。So this approach doesn't work. Maybe as proof of concept it does, but long term, no. So what can we do about it? Well, we add some rules in the front. Certain messages, especially the cold-reach messages, can be detected just by specifying the pattern, right? So, for example, the one I presented earlier: "Hey, it's been a while." If that message was just stored somewhere as a hash, you can detect that as a pattern, and potentially that's where you would make a decision even before the request goes to the public LLM. And maybe any messages related to OTPs, they can be potentially flagged as something you need to be cautious about. Well, does it work? Does this approach work? It works a little bit better, but still it doesn't really scale that well.

13:17自訓模型加公開 LLM

13:17如果你想為每一則惡意訊息手寫靜態規則跟 regex，會把自己耗盡。很難維持。解法是什麼？你做一個模型。做一個自訓的 AI 模型，處理所有基本情況。這個 AI 模型不必是大型語言模型，可以是傳統意義的機器學習模型。它能處理大多數訊息。那還要公開 LLM 嗎？事實證明要，因為有些情況模型沒那麼確定。這時公開 LLM 幫得上忙，它可以多砸一點資源、多一點訓練資料進去。And then you would be really exhausted if you try to specify static rules and regex patterns for every single malicious message. It's very hard to sustain that. So what is the solution? Well, you build a model. You build a custom-trained AI model that handles all your base case. And the AI model in this case doesn't have to be a large language model. So it can be a machine learning model in the traditional sense. And that something can handle most of the messages. Does it mean we still want public LLM to be there? Well, as it turns out, yes, because there can be some cases where the AI model is not so sure about, right? And that's when the public LLM can help, because it can throw a little bit more of the resources at it, a little bit more of the training data.

14:13不只這樣，公開 LLM 也有助於補強對訊息的判定。想像如果只有你自己的 AI 模型，也許是二元分類器，只告訴你這是詐騙或不是詐騙。或者三點輸出：詐騙、良性、或中間地帶。有了 AI，你可以再用有點像深挖的方式補強，不只看到這是詐騙，還看到為什麼、屬於哪一種分類。And not only that, public LLM is also useful for augmenting or enhancing the verdict of the message, right? So imagine if it was just your own AI model, maybe it would be a binary classifier, right, that would just tell you, oh, this is a scam or this is not a scam. Or maybe it's a three-point output where it's either a scam, benign, or maybe something in between. But with AI, you can just enhance that with a little bit like a deep-dive type of approach, where you would see not only that this is a scam, but also why, and what kind of classification is that.

15:00我們把這個概念納進去、做出來，看起來有用。封閉測試跑出來的統計：總共掃描超過 200 萬則訊息。大多數來自 Telegram，那裡有 2.7% 的訊息被標成惡意。補充一下，這些不是私訊互動裡的訊息。假設你加入大型 Telegram 群組，詐騙者傾向在那裡貼各種訊息，你收到最可疑的訊息也在那裡。Well, and yeah, we took into account that concept, we built it, and it seems to be working. And from the stats that we've seen from running a closed beta of this approach, there have been more than 2 million messages scanned in total. The majority of those were from Telegram, and that's where 2.7% of the messages were flagged as malicious. And just to note, these are messages not from the DM interactions. In case, let's say you sign up for, you join the big Telegram group, that's where all the scammers tend to post different messages, and that's where you receive the most suspicious messages.

15:48再來是 email，然後 WhatsApp。我們也試過 Slack，但是公司 Slack，那邊是零則，顯然因為那是受信任的環境，能抓到的沒那麼多。你可以看到，在這個規模、大約每天 3 萬則訊息，我描述的做法看起來運作得不錯。回到執行階段：這時再反應真的太晚了嗎？還不晚，但你這邊要多做一點工。這就到第二個解法：作業安全加固，本質上是讓筆電更安全。Next goes email and then WhatsApp, and then we also experimented with Slack, but with corporate Slack, and there were zero messages there, obviously because it's a trusted setup, so it's not as many you can have. And you can see that at this scale, at approximately 30,000 messages per day, the approach that I described seems to be working pretty well. Well, and how about going back to the execution phase? And is it really too late to react in this case? Well, it's not too late, but it will require a little bit more work from your side. That brings us to the second solution: operational security hardening, essentially making your laptop more secure.

16:40筆電作業安全檢查

16:40所以你點了連結之後，也許看到的不是被駭，而是系統沒被攻陷、一切運作正常。怎麼做到？在那之前，還有一個有意思的數字。Web3 損失越來越是作業安全問題，不只是程式碼問題。很多攻擊是因為存取控制問題，超過一半。相比之下，智慧合約漏洞只佔 12.8%。這是 Hacken 的 2025 年資料。So instead of seeing that when you follow the link, maybe you will see that you are not hacked, that your system is not compromised, and that everything is working perfectly. How to do that? Well, actually, before that, there is another interesting stat that I can share. So Web3 losses are increasingly an opsec problem, not just a code problem. And a lot of exploits were happening due to access control issues, and that's more than half, right? And just to compare, smart contract vulnerabilities only accounted for 12.8%. That's according to the 2025 data by Hacken.

17:26釣魚詐騙排第二，23.8%。釣魚詐騙幾乎就是我剛才投影片在講的事。那要怎麼把環境顧好？有一些基本檢查可以做。很多是常識，但還是得做。光用 macOS 就安全的時代，早就過去了。筆電要裝防惡意軟體，同時要有防火牆，儲存也要加密。建議不要用熱錢包，至少不要跟日常工作用同一台筆電。Phishing scams were in second place, 23.8%. And phishing scams is pretty much what I just covered in the slides so far. So what to do? What to do about securing your setup? Well, there are some basic checks that you can do. A lot of them are common sense, but nonetheless, they need to be done. The times when macOS was safe just by using macOS, those times are long gone. You need to have an anti-malware installed on your laptop, and at the same time some firewall, and also you need to make sure that the storage is encrypted as well. And it's recommended to not use a hot wallet, at least not on the same laptop as you use for regular daily work, right?

18:17你可以用硬體錢包，關於有多安全，我覺得還是有不同看法，取決於你用哪家。你也可以準備一台完全不連網的筆電，這差不多也是安全做法，前提是你在那台上也把自己該做的做完。怎麼讓這件事更系統化？至少在 Quantstamp，我們整理出 30 到 40 項作業安全檢查，這份清單大概是 2017 或 2018 年出現的。You can either use a hardware wallet, which I think there are still multiple opinions about how secure that is, right, depending on the provider that you use. But what you can also do is just have a separate laptop that is not connected to the network at all, and that is pretty much also a secure approach, as long as you are also doing your part there. And, yeah, so how do we make it more systematic, right? What we've been doing, at least at Quantstamp, we came up with a list of 30 to 40 operational security checks, and that list came up somewhere, I don't know, 2017 or 2018.

19:02我們就一直這樣做。基本上檢查公司每個人，確認有裝防毒、防火牆有開，像這樣的檢查有 30 項。包括你的社群媒體，確認處處都有雙因素驗證。好處是到了 AI 階段，很多檢查可以自動化，不必再手做。即便如此，我們內部一直在用。後來發現其他公司、其他客戶可能也有興趣。所以偶爾會跟一些客戶通話，請他們把環境加固，讓筆電更安全。And that's what we've been doing. We've been just basically checking everyone in the company, making sure that they have antivirus installed, basically firewall enabled, and there is like— there's 30 checks like that. That includes your social media, making sure that you have two-factor authentication across the board everywhere. And the good thing is that during the AI phase, a lot of these checks can be automated, so you don't have to actually do it manually. But even then, we've been using that internally. And we realized that other companies, other customers might be interested in the same thing. So we just had occasional calls with some of the customers, asking to just harden their setup, just making their laptops more secure.

19:56SOC 2 與 SEAL 標準

19:56我們就那樣做了。同時也有標準。其中一些檢查被收成標準。一個大家都知道的是 SOC 2，雖然它不是 Web3 專用。SOC 2 回答的是：你的組織有沒有被安全管理。很多新創都在做。但依我們的經驗，很多 Web3 新創沒做，因為沒有明確的合規要求。同時還有另一個標準叫 SEAL，Security Alliance，更針對一點。它回答的是：這個組織有沒有準備好處理 Web3 特有的攻擊。And we did that. And at the same time, there are standards. Some of those checks are generalized into standards. One obvious standard that you know is SOC 2, even though it's not specific to Web3. But SOC 2 answers the question whether your organization is securely managed. And that's something that a lot of startups are doing. But in our experience, a lot of Web3 startups are not doing that because there is no explicit compliance requirement to do that. At the same time, there is another standard called SEAL, Security Alliance, and that standard is a little bit more specific. It answers the question, is the organization prepared for handling Web3-specific attacks, right?

20:43這兩個標準其實不是互搶，而是互補。如果你的組織做 Web3，理想是兩個都跟。有一些差異。SOC 2 有很成熟的認證生態，會由會計師事務所認證，客戶認知度很高。我覺得很多人知道 SOC 2。SEAL 則比較新，生態還在成形，但更聚焦 Web3 相關威脅。And those are the two standards that are not really competing with one another, but they are rather complementary. It would be ideal if your organization works with Web3. It would be ideal if you follow both of them. There are some differences, right? So for SOC 2, there is the very mature attestation ecosystem, right? So you would be attested by a CPA organization, and customer recognition is very high. So I think a lot of people know about SOC 2. At the same time, SEAL is somewhat new. The ecosystem is still emerging, but at the same time, it's more focused on Web3-related threats, right?

21:31它可以引導你怎麼安全使用錢包、怎麼安全做多簽，諸如此類。Quantstamp 其實正走在成為 SEAL 認證機構之一的路上。我覺得很多公司會從中受益。喔，還有幾個有意思的案例被人拿來給我。另一個社交工程向量，是讓你去用某個合法應用的脆弱版本。第一個例子裡，駭客是要你裝假的 Teams。It can answer, it can guide you how to use wallets securely, how to do multi-signing securely, and so on. And, yeah, Quantstamp is actually on track to be one of the SEAL certifiers. So I think that's something that a lot of companies would benefit from. Oh, I had a couple more interesting cases that were also brought to me. So one other social engineering vector is making you use a vulnerable version of a legitimate application, right? So if in the first example I showed you that the hacker tried to get you to install a fake Teams app.

22:17這個案例裡，駭客試圖讓人去裝某個筆記應用非常特定的脆弱版本。那個版本能被利用。這個利用是在被社交工程的人問了之後才揭開的：「為什麼我要裝這個特定版本？為什麼不是最新版？」這就開始可疑了。這也只是再強調：更新到最新版通常是好主意。當然也有被駭客反向利用的情況。In this particular case, what happened was a hacker tried to make somebody install a very specific vulnerable version of a note-taking app, right? And that version would allow an exploit, and this exploit was uncovered after the person who was trying to get social-engineered asked, "Oh, why do I need to install this specific version? Why not the newest version?" And that's where it got suspicious, right? And yeah, this is just to highlight more that updating to the latest version is usually a good idea. There are cases when it's also misused by hackers, right?

23:02假冒交易所與事故應變

23:02所以駭客如果拿到你的 npm 憑證，就可以推一個惡意版本的套件。那有另一套預防。但同時，如果你用的函式庫、應用程式或作業系統非常非常舊，被駭的機率更高，因為利用程式已經在外面了。另一個有意思的案例是有人假冒交易所，對方其實差不多是我認識的人。發生的事是：有人多次聯絡一個在 Gemini 這家交易所開戶的人，非常像真的，還直接打電話，而這個人有把通話錄下來。So hackers, if they get their hands on your npm credentials, they can push a malicious version of a package. And there is a different prevention for that, right? But at the same time, if you're using a very, very old version of a library, application, or operating system, you are more likely to get hacked, because the exploit will be available in the wild somewhere already. Another interesting case is an exchange was impersonated, and that's actually somebody that pretty much, yeah, I know personally. So what happened is, yeah, there were multiple attempts to reach out to somebody who has an account at Gemini, the exchange, and it was extremely convincing, with phone calls reaching out to the person directly, and the person managed to record the phone calls.

24:05然後我同事把那通錄音丟進 LLM，用分析語音的功能，偵測結果是 99% 機率是詐騙。這是這個做法一個很有意思的應用：LLM 可以幫你偵測惡意活動。更別說那封 email 一收到，就被 Common Defense 標出來了。最後，駭客真的發生了怎麼辦？所有防線都破了，然後呢？And then my colleague sent that phone call to an LLM, basically to the feature that is using LLM for analyzing voice, and it was detected to be a 99% probability that this is a scam, right? And that was a very interesting application of this particular approach, that, you know, LLM can help you detect malicious activity. And not to mention that the email itself was flagged by Common Defense as well whenever it was received. Well, and finally, what to do when the hack actually happened, right? So if all the layers of defense failed, what to do then?

24:50這就到第三個解法：事故應變。怎麼更快調查、更快從問題裡恢復。很多聽起來像常識，但你第一次碰上會非常茫然，不知道該做什麼。面前只要有一份步驟清單，也許八到十步，或有人熟悉這些步驟，就很有幫助。你先做什麼？先止血。把裝置斷開。我覺得重要的是把裝置從網路斷開，因為比較可能發生的是這個。Well, that brings us to the third solution. It's incident response. How to investigate and recover quicker from the issue. And a lot of it might sound like common sense, but the first time you are in this situation, you will be very lost. You don't know what to do, right? And having just a list of steps in front of you, a list of maybe eight or ten different steps, or somebody who is familiar with those steps, would go a long way. So what do you do? You first try to stop the bleeding, right? You disconnect the device. I think the important part will be disconnecting the device from the network, because that's what would happen more likely.

25:36你需要保全證據。要截圖、記下時間戳跟雜湊等等。這裡的建議還有：筆電保持開機，斷網但不要關機，這樣如果有惡意行程，會留在記憶體裡。你要通知對的人。通知安全團隊、領導層，也許還有合作對象來幫忙。最難的是，如果資金被偷，要追蹤並救回。成功率高低不一，但有些情況，如果你做對步驟，非常非常有可能。重點是：這個領域有一個很大、還在長大的社交工程問題。You would need to preserve evidence. You need to take screenshots and record timestamps and hashes and so on. And the advice here is also to keep your laptop on, disconnected from the network, but keep it on so that if there is a malicious process there, it stays in the memory. You need to alert the right people. You need to alert your security team, leadership, and maybe anyone you are in partnership with to help you with that. And the hardest part will be to track and recover the funds if they are stolen. That has a varying level of success, but in some cases it is very, very possible if you do the right steps. Well, the takeaway is we have the big and ever-growing social engineering problem in the space.

26:28三層解法：偵測、加固、應變

26:28AI 讓它更糟，但 AI 也能讓它更好。我們走的三個解法是：統一偵測，本質上就是檢查通訊有沒有惡意活動的平台；作業安全加固，把環境顧好；以及事故應變。謝謝大家，很榮幸在這裡講。請到 commondefense.ai 看看。非常感謝。AI makes it worse, but also AI can make it better. And the three solutions we go with is unified detection, essentially the platform for checking the communications for malicious activity. You have OPSEC hardening, which is about securing your environment, and incident response. Thanks, everyone. It was a great pleasure talking, speaking here. And yeah, go to commondefense.ai. Check us out. Thank you so much.