AI 會破解加密還是讓它更安全？Pantera、Chainalysis、Agora 對談

0:00開場與嘉賓介紹

0:00大家好，感謝大家來參加這場討論。我們今天要聊的主題是 AI 帶來的巨大浪潮，這感覺有點像早期的加密貨幣時代，既新鮮又充滿未知。很高興邀請到幾位重量級嘉賓一起參與。先請大家自我介紹一下，然後我們就直接進入主題吧。Well, thank you everybody for coming. Excited for this panel. Obviously, there is a massive tsunami coming in the technology field, which is AI. It kind of feels like crypto in some sense, where it feels very early, but kind of unknown. And so great to have these esteemed guests join me. Maybe to start, we can just do a quick round of intros and then jump into things. Sure.

0:34大家好，我是 Cosmo，負責 Pantera Capital 的投資策略。我們是一家科技投資公司，從早期創投到公開市場都有涉獵。接下來請 Drake 介紹。Hey everyone, my name is Cosmo. I lead investment strategy at Pantera Capital. Pantera is a tech investment firm. We have early stage venture all the way through public markets practices.

0:46我是 Drake Evans，Agora 的共同創辦人暨技術長。Agora 提供網路原生貨幣的一站式入口，我們有自己的穩定幣 AUSD，也提供白標發行服務，未來還會有銀行執照、託管、支付與美元帳戶等功能，還有錢包，真的是一條龍服務。My name is Drake Evans. I'm the CTO and co-founder of Agora. Agora provides a single and final entry point for internet-native money. We have a stablecoin, AUSD. We also do white-label issuance, and soon, with the finalization of our bank charter, custody, payments, U.S. dollar bank accounts. We also do wallets, really the one-stop shop.

1:09大家好，我是 Jonathan，Chainalysis 的共同創辦人暨執行長。我們為產業、監管單位和政府機構提供情報服務，協助保護加密產業並讓它更強大。Hi, my name is Jonathan. I'm co-founder and CEO of Chainalysis. We provide all of the intelligence for the industry and regulators and government agencies that want to be able to actually protect crypto and make it stronger.

1:25我是 Logan，負責 Frictionless Capital，是一家早期創投基金。那我們就直接切入主題吧。Perfect. And I'm Logan. I lead Frictionless Capital, early-stage venture fund. Maybe starting with you, Jonathan.

1:33AI 與加密安全攻防

1:33Jonathan，想請你聊聊鏈上安全。我們剛剛在後台有聊到，現在 AI 和加密安全有點像貓捉老鼠的遊戲，AI 一直在試圖找漏洞，而我們的智慧合約也在不斷進化。你怎麼看現在的市場情勢？Kind of diving deep onto on-chain, we were talking right before we walked on stage on the encryption side. It's a little bit of a cat and mouse game between AI kind of trying to poke holes in code versus what we have with smart contracts. What's kind of your current read of the market?

1:54我們現在已經看到 AI 被用來發動攻擊，針對協議、區塊鏈甚至錢包。作為防守方，最大的挑戰是目前防守還是需要人類參與，但攻擊方未必有真人在環節裡。這種不對稱會造成運作速度上的問題，這也是我們這個產業會遇到的困難。We start to see is that, you know, AI is clearly being used in exploits that are being leveraged on, you know, attacking protocols and chains and, you know, even wallets. And I think that, you know, the difficult thing that we have to do as defenders is that there is always going to be human in the loop currently on defense. Whereas there is not necessarily human in the loop on attack. And when you have that type of asymmetry, you have issues on, like, what speed you can operate at. And I think that this is where you start to see problems occurring in the sector.

2:35不過話說回來，今年最大的竊案其實還不是什麼 AI 機器人大軍入侵，而是基本的資安問題，比如私鑰外洩等傳統漏洞，這些還是最常見的問題。That being said, the biggest, you know, thefts of this year are still not. Like agent swarms hacking into people's stuff, it's actually just like getting the basic cybersecurity stuff right. And I think that, you know, private key compromise and other things that we've seen, you know, are just traditional cybersecurity failures rather than, you know, completely brand new exploits.

3:00AI 會打破加密嗎？密碼學危機與反思

3:00剛剛我們也有聊到 Justin Drake 最近在 X 上的熱門貼文，討論 AI 可能對加密市場帶來的毀滅性影響，甚至可能破解密碼學。Drake，你怎麼看這件事？And Drake, we were talking earlier again about Justin Drake's Twitter post. If you have not seen it, it got fairly viral on Twitter and X about kind of impending doom of AI on crypto markets and potentially breaking cryptography. And you also had some very interesting thoughts here.

3:19最近 24 小時，X 上討論很多，像是一些老數學難題被解決，或是大家以為的障礙被突破。人類通常會找優雅的解法，但 AI 可能會用很粗糙的方式解決問題。大實驗室最近公布了上百個被破解的數學證明，但沒有一個跟密碼學有關，這其實很可疑。如果真的有機會用不優雅的方式大幅縮短破解非對稱加密的週期，我們就要開始準備了。Yeah, I mean, I think if you've been, you know, looking at Twitter the last 24 hours, there's been a lot of discussion about the, you know, a bunch of old, old math proofs or math problems being solved, or, you know, these sort of assumptions about the barriers or these asymptotes being broken in, you know. What people are calling, like, an inelegant solution, right? I think humans have a tendency to find elegant solutions because of small context window and other things. And so then the thesis goes, okay, you know, the big labs have posted a hundred math proofs that they broke in the last few months. Zero of them are related to cryptography. That's a bit suspicious. So. You know, if there's a non-zero chance that there is an inelegant solution to significantly shorten the cycle of breaking what I would consider most of the, you know, asymmetric cryptography algorithms as elegant, you know, we should be preparing for that.

4:18當然也有人認為 AI 會幫我們找到更好的加密演算法。Cosmo，Pantera 作為 OG 級的加密基金，你們經歷過市場高低起伏，現在你怎麼看這個新時代？I think there's been a bit of a—the counterargument, of course, is that AI will help us find better algorithms for the future. And Cosmo, Pantera has been one of the OG, OG crypto funds. You guys have seen a lot of high highs, some low lows. What's kind of your current market take, more from an investor standpoint, about kind of maybe the shifting landscape as there is new paradigms?

4:53AI 對產業與區塊鏈的雙向影響

4:53我認為最大的典範轉移就是 AI 對各行各業的影響。每個人都應該思考自己的技能怎麼搭配這波 AI 浪潮。作為區塊鏈投資人，我們會想區塊鏈怎麼加速 AI 創新，反過來 AI 又會怎麼影響區塊鏈普及。這可能會打破加密，也可能加速加密發展。Well, I think the biggest paradigm is clearly the impact that AI has on any industry. And so, as I think as any unique human, I'd encourage you to think about how you and your specific skill set or toolkit can be used to accelerate what is one of the biggest tsunamis, as you described, Logan, which is the influx of AI innovation and its penetration of society. And so us as blockchain investors, naturally then we are thinking about how does blockchain play a role in accelerating AI innovation, or on the flip side, how does AI impact blockchain adoption? One of which is maybe breaking crypto, but the other is accelerating crypto.

5:32這可能有點爭議，但我真的認為智慧合約被駭的時代快結束了，未來三到六個月內，能被駭的都會被駭光，之後新發佈的都會經過完整滲透測試。換句話說，DeFi 被駭的問題快要落幕了。但 Jonathan 剛剛說的人性漏洞還是核心問題，這才是真正的挑戰。You know, maybe a controversial take, and I'm sure there's holes with this, but I do believe we're nearing the end of smart contract hacks. Like, the era of smart contract hacks, I think, will be over in the next three to six months, because anything that can be hacked will have been hacked by then, and that anything going forward will have been, you know, fully penetration tested before it's been released. And so. You know, the optimist view is that we're almost done with DeFi hacks. Now, that doesn't stop anything Jonathan said about human hacks, right, which is really the core of the problem.

6:07沒錯，尤其是傳統軟體和智慧合約相比，智慧合約的攻擊面其實小很多。我很樂觀，防禦工具越來越強，未來幾乎可以百分百確保智慧合約安全。傳統軟體就難多了，因為一直在變動，複雜度也高很多。I think that's absolutely right, especially, you know, if you look at sort of traditional software versus smart contracts, the surface area for smart contracts is just generally significantly lower. And I'm, you know, very optimistic that the defensive tools are getting better and better to the point that you can have as close as 100% certainty a smart contract secure. Traditional software is a bit harder, one, because it's constantly changing and there's typically a lot more complexity and surface area.

6:35我想反駁一下，其實沒有什麼軟體是完全安全的。不論是 AI 寫的還是人寫的，總會有失敗的地方。雖然防禦工具進步很多，專案營運也越來越專業，但我們不能以為系統會絕對安全，這是不現實的。I'll take the other side. Okay. I mean, I think... No, I think that, like, there is no such thing as, like, secure, like completely secure software. And I think that we're at this point where not all of those languages are necessarily, like, able to produce secure smart contracts. And I think that at the end of the day, whether it's agents writing them or whether it's humans writing them, there's going to be failure modes inside of that. I think that the problem—Made change, and I think you're both right. The defense against these things has got a lot better. You know, the operations in projects is getting a lot more sophisticated. And yes, like lots of things that were going to get hacked have been hacked. But I think that, you know, fundamentally, I don't think we can live in a world where we think that there are going to be fundamentally secure systems that everyone can just rely on.

7:32防禦策略：形式驗證與詐騙對抗

7:32假設未來還是會有 AI 利用漏洞發動攻擊，我們該怎麼防禦？是要靠形式驗證，還是 AI 防禦？你們怎麼看？So assuming that maybe we aren't at the end of finding all the bugs and that there is still going to be some potential hacks or bugs that are exploited with AI over time, what's the best way to try to protect about that? Is it formal verification? Is it kind of AI defense, where do you think that kind of leads us?

7:53我認為未來會有更多形式驗證方法，AI 也會降低這些難學語言的門檻，讓更多人能用。其實產業最大挑戰不是智慧合約漏洞，而是釣魚和詐騙，這是全球問題，不只是加密圈。現在 AI 生成的影像、聲音、文字都很逼真，大家很難分辨是不是被騙。Yeah, I think that maybe one of the things that I think, you know, will be necessary, I think there will be more formal verification methods in general, and I think that AI actually will lower the barrier to entry for some of those programming languages that have historically been very difficult to get people to adopt in. Even in the crypto industry. But actually, I think that the biggest challenge that we're going to have as an industry is actually not in the smart contract vulnerability landscape, much more in, like, phishing and scams. Like, this is something that I don't think is even, like, a crypto problem. Like, this is just a global problem. Everyone here has, like, experienced unbelievably good—Video, audio, text-based AI that is so believable that will morph someone's ability to understand whether they're being scammed or not.

8:50最好的防禦其實是主動出擊。我們不能只靠訓練人類辨識 AI 詐騙，必須用 AI 代理人假裝受害者，主動去釣出詐騙集團的金融基礎設施，這樣才能真正解決問題。Chainalysis 這兩年就在做這件事，用代理人群組大規模跟詐騙犯互動，找出他們的運作基礎設施，這才是有效的預防措施。And I think that the best form of defense to the question is actually offense. And I think that you can't think that we're going to train humans. To detect AI scamming them from things, you're going to have to go meet scammers at scale using agents pretending to be victims to be able to actually uncover the financial infrastructure that they're running on in order to best actually address this problem. And this is what we've been doing at Chainalysis for the last two years, is like we've been building. Agent swarms effectively to go actually and talk to scammers at scale to be able to solicit all of their infrastructure that they're running on. And that's like the only way that you can get, like, real preventative, you know, care against these types of threats.

9:43現在已經不是單純拖延詐騙犯時間的時代了，因為他們的規模可以無限擴張。我們要做的是找出他們真正依賴的金融基礎設施，這才是稀缺資源。DDoSing scammers. Not just DDoS, but actually, like, I think that this is the thing, is like—It used to be the case that you could, like, waste a scammer's time, and then they wouldn't be able to call Cosmo. But, like, now, like, they've got infinite capacity. So what you have to go and get is their, like, financial infrastructure that they're relying on, which is scarce as a resource.

10:09DeFi、投資觀點與量子威脅

10:09Cosmo，從投資人角度來看，DeFi 一直是區塊鏈的核心基礎設施。這些安全問題會不會影響你們對這些資產的投資信心？還是你覺得 DeFi 的漏洞會在未來半年到一年內被解決？Cosmo, from again maybe more the investor lens, DeFi was really one of the core primitives and continues to be one of the core primitives within blockchains and kind of these capital markets that we're building. Does it change kind of your view on the investability of any of these assets? Or again, do you think all these DeFi bugs just get solved within the next—Six to 12 months.

10:35這確實讓我們在資本配置上會更謹慎，因為很多東西還沒完全安全。不過市場其實很有韌性，很多中心化機構都經歷過被駭還是活下來了。市場看重的是能不能提供穩定可靠的體驗，而不是百分百沒出過事。我們會對新協議或沒經過戰場考驗的東西更謹慎，但也不能因噎廢食。I would say it definitely gives us some pause about how we should be allocating capital in today's world. There's no doubt that a lot of this stuff is, you know, not fully secure as has been described. I think there's a—You know, there's a risk tolerance that we're okay with. And there's, by the way, like a lot of things can survive hacks. I mean, almost all of the centralized entities have survived hacks, right? And so there's a little bit of like the market kind of sees through it because at the end of the day, a business is more than just like, has it been secure 100% of the time. It's more that does it provide a consistent, reliable experience than like five nines maybe. And so, you know, there is risk. There's definitely increased risk, and we're definitely more, just more cautious, more curious around, like, net new protocols or things that haven't been as battle-tested. But, you know, there's always stuff to do, and you can't be scared of your own shadow.

11:34我們也聊到協議的「lindiness」（存活時間越長越安全），其實再怎麼壓力測試都不如時間考驗。現在除了 AI，大家也開始討論量子電腦的威脅。Drake，你怎麼看量子對加密的影響？In Drake, we are also talking about kind of the lindiness of some protocols. It's really like you can stress test them as much as you want, but really you just kind of need time in the market. But there's also kind of this, like, third factor that people have been talking about on the timeline, which is quantum. And kind of not only do you have AI, but potentially you have quantum coming up. And what are your kind of thoughts about? Quantum as well.

11:59我覺得 AI 會加速很多既有趨勢，包括代幣化和量子破解的進展。原本大家以為量子電腦要十年二十年才會威脅到加密，但現在看來可能會更快。像 Vitalik 今天也有發文，說過去幾年以來 Ethereum 的規劃其實都在為這種未來做準備。那些有前瞻性、假設事情會比預期更快發生的團隊會表現得更好。I think it's probably very related to what we were talking about earlier, and just that AI—and this is, like, I think a general thing that will affect crypto and all industries—is AI is going to accelerate a lot of existing trends. I think it will accelerate tokenization. It'll also accelerate our path to breaking quantum, which I think most people had accepted as something that would happen in the next decade or two. Even conservatively. And so, you know, I think it's funny, Vitalik posted today about kind of like how all of the plans they've been implementing over the last few years, like, play into this kind of roadmap. And I think, you know, teams who have that forward-looking kind of mindset and organizations that sort of operate on the assumption that things will move faster than they expected will do very well.

12:48過去一年 AI 能力幾乎是指數成長，2023 年還有人說 AI 只是像昆蟲、狗、嬰兒，結果一眨眼就超越人類了。現在唯一能跟上這個浪潮的方法，就是假設創新會越來越快，時程會被壓縮。If you look at even just AI over the last year, you've seen this just like what feels like an exponential growth in capability. But there were people in 2023 saying, this is, you know, you've seen the meme graph, right, where it feels like it's just, it's like the intelligence of an insect and then a dog and then a baby, and in a blink of an eye it'll be, you know. Better than a human. And so I think we're seeing that play out, and making the bet that timelines will compress, innovation will go faster is the only way to stay ahead of it.

13:20AI 加速創新與產業樂觀展望

13:20這場討論好像有點悲觀，AI 好像會帶來很多麻煩。不過想問問大家，現在有什麼讓你們感到興奮的事？你們都還在這個產業，應該還是有樂觀的理由吧？So this panel has been a little bit doom and gloom since that AI can really kind of throw a wrench in things. But maybe an open question to the panel is: what are the things that are getting you guys excited now? Crypto markets have matured. We've run a lot of experimentations. Obviously, you guys each are still in the industry. I'm assuming because there is some optimism. What are the things that currently excite you?

13:45我覺得最讓人興奮的就是所有既有趨勢都會被 AI 加速。AI 會加速代幣化，雖然有很多不確定性，但它確實降低了創業門檻，讓大家更容易做新事、創新，未來幾年一定會有很多很酷的東西出現。I think it's really just the acceleration of all of these existing trends. I think that AI will accelerate tokenization. I think the big thing is AI is, you know, there's a lot of uncertainty, but it's—Definitely lowering the barriers to doing more stuff broadly, whether that's starting a new business, you know, in some cases attacking software. But, like, in general, like, people are going to be doing more stuff. That means more innovation. That means more creativity. That means, you know, we're going to see really cool stuff, I think, over the next few years.

14:18我們現在已經在大規模用 AI 對抗產業威脅，特別是在詐騙和非法活動方面，AI 幫助我們搶先一步。下一步是讓所有客戶都能用 AI 來調查、合規，節省時間和成本，專注於業務成長。AI 也會縮短客戶回應時間，讓加密產品更普及。I think that, like, from our perspective. Firstly, like we're using AI now at scale to, like, you know, tackle some of the big threats to the industry, and, like, it's actually helped us get ahead of some of those threats, you know, particularly on the scam side, but also, you know, the various different, you know, types of illicit activity that we—The retracker chain analysis, like we're able to leverage it to be able to protect the industry at greater scale. I think the next chapter of this is actually helping all of our customers leverage AI to be able to perform investigations, to do compliance more efficiently, to really, you know, save on, you know, the time and expense that goes into, you know, running, you know, operations at different. Businesses so that people can be focused on, like, growing their business and doing better for their customers. I think that, you know, AI will shorten the latency time for response for customers. It's going to increase, you know, how accessible crypto products are in general for everyone.

15:19我們在 Chainalysis 看到亞太地區的支付創新比其他地方多很多，很多實體商品和服務都用穩定幣作為底層支付。AI 和加密在推論市場的交集也越來越大，像是用加密貨幣支付 AI 推論服務，這是我們現在常看到的趨勢。這也是創新金融服務和穩定幣銀行爆發的好時機。I think that the things that we're seeing that are exciting at Chainalysis is, you know, coming to this region. There is just so much more happening on payments in APAC than, you know, in other parts of the world. There's lots of stuff happening for the real goods and services that are moving based on stablecoin rails. There is also, like, a large intersection of, you know, AI and crypto in, you know, the inference market and—Seeing your inference being paid for using crypto is something that we start to see a lot of at Chainalysis. And, you know, I would say that, you know, it's a great moment for people to innovate on new types of financial services based on stablecoin banks that we heard from the previous panel. I think that this is something that, you know, has exploded over the last—Yeah, year since Genius in the States, and I think that that's only something that will continue to accelerate.

16:21我對現在區塊鏈協議的投資風險報酬比非常看好，因為 AI 會大幅加速鏈上活動，尤其是 agentic activity（AI 代理人之間的互動）。代理人之間需要交易，傳統金融流程又慢又貴，區塊鏈錢包 24/7、全球即時又便宜，未來一定會選擇區塊鏈。AI 經濟和代理人經濟的成長會帶動整個區塊鏈產業，這是很大的順風。So I think that I'm, like, very bullish on the ways in which we're now tapping into real economic activity running on blockchain rails. It's what excites us the most at Chainalysis, is seeing those flows. And actually, like, I think that, you know, it's an awesome moment to be a builder. Like, it's an awesome moment to be able to experiment, and, like, that experimentation is what crypto is actually fundamentally all about. I think the risk-reward for investing in blockchain protocols today is the best that it has been in many, many years. And the reason I say that is precisely because I think AI is going to be such a huge accelerator of on-chain activity, if you think, and largely because of agentic activity. When you think about, like, what agents are, agents will want to be able to transact with each other. They want to, and when they transact with each other, you know, they'll have to find some sort of transaction rails, and they have option A, option B. Option A is—Walk into a physical branch, present two forms of ID, wait a few days to get a bank account, and then have the privilege of sending money five days a week, eight hours a day, or spin a blockchain wallet that's 24/7, global, cheap, fraction of a penny, anywhere, instantly. They're obviously going to choose option B. And so, as we see, to the extent you believe at all in the growth of AI and the growth of the agentic economy and the growth of agentic intermediate commerce, like it's obviously going to accelerate everything across blockchain, and that is such a massive tailwind.

18:07現在 AI 相關股票漲很多，區塊鏈反而還比高點跌了四五成，但其實兩者都受惠於同樣的長期趨勢。對我來說，這是很棒的投資機會。And at the same time, you know, you see there's plenty of ways to invest in AI, and we see all seen AI stocks and AI-related things go up a ton. Definitely not saying they are necessarily overvalued by any means, but they've gone up a lot. Relative to blockchain, which is down 40, 50% from the highs, even though they're riding along the same secular tailwinds. And so to me, that seems like an incredible, incredible risk-reward.

18:19AI x 區塊鏈的投資與創新機會

18:19很多人想知道怎麼押注加密、AI 或兩者交集。Cosmo，你怎麼看 agentic 經濟的投資機會？Maybe back to you, Cosmo. And I think right now there's a lot of people probably in the audience are trying to learn about what is the best way to express going long crypto, going long AI, the possible intersection. Is it—On the agent side, how do you kind of see that?

18:35這時候就要從宏觀拉回具體執行。Chainalysis 和 Agora 的團隊執行力都很強，大家可以去看看他們的產品（不是業配）。我們最看好的區塊鏈協議有 DePIN（去中心化實體基礎設施）相關的，因為 AI 發展的瓶頸在於能源、算力、資料。像 Grass 做資料、Glow 做能源和算力，還有很多人在做分散式推論和訓練。分散式訓練比較難，但分散式推論很有機會。Well, this is where, like, you have to drop down from generalisms to actual ideas that are interesting, because there are plenty of ideas that are interesting, but it's execution. Name of the game is execution. I mean, the guys at Chainalysis and Agoric have done an excellent job in execution. You check out both their products. Not a plug. I'd say, like, within blockchain protocols, some of the protocols we find most exciting are anything across, you know, DePIN protocols I think are going to be very interesting in a world of AI. It's very clear that some of the bottlenecks to AI adoption are energy, compute, data. There are plenty of DePIN networks across energy, compute, data, right? There's Grass doing amazing things across data. There's Glow and others doing amazing things across energy, compute. There's all sorts of ways that people are thinking about distributed inference and distributed training. Distributed training may be tougher, but distributed inference. And so that's, like, super exciting.

19:33我們最近也投資了 Worldcoin，proof of human（人類證明）這個概念現在變得很重要。未來大家在線上互動越來越多，辨識誰是真人會越來越有價值。Or whether across identity, we just led a round in Worldcoin, which has, you know, been this thing that sort of—Been kind of a crazy dream and a crazy promise for a long time, but this concept of proof of human is actually really relevant today. The idea that you and I are going to do more online, and that when we're in the digital world, the desire to know who is truly human will only increase over time, and that identifier will become very valuable over time is massive.

19:58AI 實際應用與組織內部創新

19:58Drake，你們 Agora 怎麼把 AI 融入產品和內部流程？And Drake, how are you guys incorporating kind of AI within Agora?

20:02我們有兩個層面：一是內部怎麼用 AI，二是 AI 會怎麼影響客戶。對客戶來說，未來競爭會更激烈，消費者不再像以前那麼黏著，產品和經濟效益要最好，還要降低摩擦。像選擇錢包比開銀行戶頭簡單很多，因為價值兌現的摩擦很低，這會改變大家消費產品的方式。我們專注於降低使用者獲得價值的門檻，並提供基礎模組，讓大家能自己組合應用，不用再靠 SaaS 幫你整合。I mean, I think there's sort of two ways to think about it. The first is how we're using AI internally, and then the second is what do we believe about what AI will do to our customer base. I think, you know, on the customer base side, we believe that you're going to have to compete. You know, consumers are no longer sticky in the same sense, right? There's been a lot of talk about this with bank deposits, but But it's really, you're going to have to have the best product and the best economics and have the aggregation and lower that friction. You know, we were just talking about how, you know, choosing a wallet over a bank is so easy, partially because the friction to value is so fast, right? There's so little friction to get the value, I think. So that's going to change how people think about consuming products. And so we're focused very much on How do we lower the friction to first sort of value, and then how do we provide those base building blocks? Because that's the other piece that is going to change, is that people are going to start building. They want access to the metal sooner and sooner, and the deeper infrastructure, because now, you know, this whole middle ground where you paid a SaaS product to kind of combine a bunch of stuff for you, you can do the combinations yourself. And so now you want the Legos.

21:09內部來說，我們把每個人類決策都當成自動化機會。只要有流程，就可以打造專屬 AI 工具，先給建議，信心夠高時就自動化。像我們的使用者審核流程，AI 會幫忙整理申請人資料、下一步建議，團隊評分後再把資料累積成訓練集，未來就能自動化更多。這套方法也適用於安全、合規、法務等部門，每個人都想要專屬的 AI 工具來包裝自己的工作流程，最後就能自動化。And then on the internal side, I think, you know, we view every single human decision as an opportunity, right? Anytime that a human needs to do something and is a process, like that's an opportunity to build a custom harness to help them with their workflow that over time, you know, starts off providing recommendations, then, you know. That high confidence is automating things. And so, you know, the first place we really pushed on this is in our onboarding, where, you know, when our onboarding team wakes up in the morning, they have, you know, a brief of everyone who's applied, what potential next steps might be, what things that might flag that, and then they can sort of rate those. And then you collect that data that becomes training data. Eventually it's large enough, you can do post-training. So really, and that's true for security, compliance, for legal. Like everyone in our company wants a custom harness that wraps their workflow, that becomes a structured input for what they're doing that can then automate it.

22:08AI 代理人防禦模型與未來展望

22:08Jonathan，你剛剛提到 agent swarm（代理人群組）這個概念很有趣，能不能再深入分享一下？感覺就像有上萬個 AI 代理人在防守智慧合約，抵禦攻擊。And Jonathan, you talked earlier about the agent swarm. I think it's a very interesting idea. Just picturing it in my mind as you have these giant agent battles going back and forth, and you're putting up your 10,000 agents defending the wall and making sure smart contracts don't get exploited. Maybe, like, dive slightly deeper into that, because I do think it's a very interesting kind of mental model of defending the smart contracts, and you guys are really—Very close to the metal, what you're doing.

22:37我們每天不只看智慧合約，還會追蹤全球新註冊的網站，分析哪些交易或網站在針對智慧合約或投資人發動詐騙。重點是要即時回應、蒐集證據，讓人們能依賴這些事實。AI 可以把過去人類做的事自動化，並把證據存下來，提供給需要的人。這樣我們就能從機率模型轉向可行動的事實，這是未來防禦者的重點。Yeah, I mean, maybe to give you a sense of, like, what we look at on a daily basis. I mean, we look at not just smart contracts, but we also look at, like, every website that's registered in the world every day. And, you know, at a scale where we're trying to map out, you know, what are the transactions that are targeting, you know, these different smart contracts or—You know, how are these websites targeting, you know, investors or people to, you know, try and get, you know, scams off the ground? And all of that is like, how do you respond to that instantaneously? How do you capture evidence that someone can then rely on? And I think that AI is there to, you know, capture what humans were doing before. And you can store it as evidence and serve it to people as facts. And I think that this is where, you know, we get away from some of the probabilistic, stochastic models out there, and you still can, like, use it as a tool in the toolkit, but produce, like, really reliable evidence. And I think that this is where. You know, the future of Defenders really happens is like translating all of this, like agentic work, which is probabilistic in nature, to like hardcore facts that you can act on. And that's a lot of what we've been spending the last couple of years thinking deeply about.

23:58太棒了，時間差不多，請大家給這幾位嘉賓熱烈掌聲！Awesome. Well, I think we're at time, so please help me in giving all these gentlemen a round of applause.