--- title: "人工智能代理協助揭露加密貨幣犯罪活動" type: "News" locale: "zh-HK" url: "https://longbridge.com/zh-HK/news/287197438.md" description: "AI 代理正在增強對加密犯罪的調查，利用 AI 進行的詐騙每次操作平均產生 320 萬美元的收益。像 Chainalysis 和 Elliptic 這樣的公司正在利用 AI 追蹤可疑交易並與詐騙者互動，提高金融犯罪檢測的效率。這項技術使得非法活動的識別速度更快，同時也為監管機構提供了反擊這些犯罪的機會。犯罪分子與調查人員之間的持續鬥爭仍在繼續，雙方都在採用先進技術" datetime: "2026-05-21T10:05:41.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/287197438.md) - [en](https://longbridge.com/en/news/287197438.md) - [zh-HK](https://longbridge.com/zh-HK/news/287197438.md) --- # 人工智能代理協助揭露加密貨幣犯罪活動 - **Key insight:** AI agents are aiding crypto crime and fraud investigations. - **What's at stake:** Scams using AI generate an average of $3.2 million per operation, compared to $719,000 for those that don't, according to Chainalysis' 2026 Crypto Crime Report. Agentic AI is helping crypto crime investigators talk with fraudsters and understand what they're doing. It's helping research the circumstances around suspicious crypto transactions. And it's helping inexperienced financial crime investigators understand the cases to which they're assigned. These AI assists come at a time when more banks are thinking about offering crypto trading to their customers. It also comes at a time when scammers and criminals are increasingly using AI in their work. Scams using AI generate an average of $3.2 million per operation, compared to $719,000 for those that don't, according to crypto monitoring firm Chainalysis' 2026 Crypto Crime Report. That nearly fivefold increase in revenue per scam is driven by AI's ability to automate impersonation — a tactic that saw 1,400% growth this year, according to the company. "Bad actors have always been the earliest adopters of any new technology, and AI is no different from that," said Simone Maini, president of crypto monitoring firm Elliptic. "So we're certainly seeing them look at, how can we get more scale, how can we probe more potential vulnerabilities. It's also reducing the cost threshold, so it means it's a little bit easier to try to penetrate different things at the same time." But crypto watchdogs "also have a huge opportunity to leverage the same technology and the same techniques against them," Maini said. "So I don't really see this as anything other than a continuation of the kind of cat-and-mouse game that we've been playing as an industry when it comes to financial crime detection." Such efforts are effective, according to Felix Shipkevich, law professor at Hofstra University and principal at Shipkevich PLLC. "AI is used by banks and financial institutions today to identify suspicious transaction patterns, wallet behavior, clustering, and links to known illicit actors faster than manual review alone, but of course it is still a tool that requires human compliance judgment," he said. Chainalysis and Elliptic, the companies most banks and DeFi companies turn to for crypto transaction monitoring, shared how they're using AI in this work. ## Chatting with criminals Chainalysis tracks more than 400 blockchain networks, including the bitcoin blockchain, ethereum and Solana. "We continue to expand on that based on where the economic significant activity is, and the amount that we track covers over 99% of the total volume," Chainalysis CEO Jonathan Levin told American Banker. To gain visibility into crypto transactions, Chainalysis collects data on all publicly available blockchain transactions, often by conducting its own transactions on those chains. It determines which transactions are coming from the same wallet. It then connects those digital asset wallets to people and entities. The company watches different threat types. For instance, it follows the movements of ransomware groups using crypto. "We look at their laundering of victim proceeds," Levin said. "We have collection strategies that look at scammers." The company passes everything it learns on to its customers, to help them with their fraud detection and cybersecurity defense efforts. Chainalysis has recently begun using agentic AI to create fake fraud victims that chat with scammers. The AI agents find out all the behaviors of the scammers and criminals and pass them along to customers, to help them make risk decisions, Levin said. AI agents haven't replaced human investigators, but help them scale their work, he said. "We spent 12 years using manual effort to make transactions to interact with threat actors, and we still very much believe in the ability to do that manually," Levin said. "But now we augment that work with the ability for our analysts to orchestrate AI agents that can do some of those interactions, that can do some of the research work, so the analyst is validating the work of those agents, and those agents are able to have conversations with threat actors at scale." Chainalysis also gathers information about these criminals' activity on traditional payment networks. "What we realized talking to these scammers and asking where they want us to send the money, sometimes they said crypto, but more often than not, they said, send it to my Zelle account, or my Cash App account, or my PayPal account," Levin said. The company provides all this intel to financial institutions, so that they can prevent victims from sending money to scammers. ## Doing research for financial crime investigators Elliptic, which has 750 clients, tracks more than 65 blockchains and monitors activity across multiple chains. "When someone tries to take something that used to be, let's say, bitcoin, and then they convert it and send it through an ethereum bridge, we can still see those transfers, and that's really important if you're going to have a full, comprehensive view of what's happening on a trade," Maini said. Sometimes the company runs nodes on these blockchains to see the transactions running on them. It also performs its own crypto transactions and runs heuristics and machine learning models to understand and fingerprint sketchy behavior. Elliptic doesn't try to unmask blockchain users, but it looks at where transactions come from. "We're trying to understand when is something a legitimate, regulated exchange like Coinbase, and when is it actually a mixer that has been, let's say, created and state sponsored by the Russian government, so that when one of our customers is receiving funds, they've got the confidence to know, that's come from Coinbase, or that's come from this mixer that looks like it's being used to launder funds," Maini said. It aims to help customers make decisions about which transactions to allow and which to put in a queue for investigation. AI comes into this work in a few ways. Elliptic designs AI agents to do some of the repetitive work performed by compliance teams to resolve alert queues so that the compliance specialists can focus on cases that are ambiguous or highly risky, Maini said. This is similar to the way many anti-money-laundering software companies are using agentic AI to help investigate cases. "It's interrogating, why has this \[transaction\] come back \[with a high risk rating\]? Is that consistent with any other activities, or does this look like it might be a false positive?" Maini said. "It's making a judgment on those things for the human to then review. So it's hugely speeding up and therefore increasing the efficiency of the overall compliance operation." Elliptic also uses AI to provide real-time natural language context, as a help to inexperienced digital asset compliance teams. "One of the real constraints that digital asset teams and banks in particular can face is there aren't enough digital asset compliance specialists out there," Maini said. "So getting new people up to speed can be a real barrier." With the context provided, when a new person looks at a case, they don't have to be a subject matter expert. "We can just provide them with that natural language explanation of, OK, this is what you're looking at," Maini said. "This is why it's scored at eight." For instance, if a customer is trying to withdraw money from Garantex, the Russian crypto exchange, a newbie might not know what that is. The natural language context explains that it's an exchange that facilitates money laundering and was taken down by the U.S. Secret Service last year. "It immediately gives them that context that allows them to make a much quicker decision such as, this actually does look pretty serious, and needs to go into an escalation," Maini said. Shipkevich noted that while banks can use companies like Chainalysis and Elliptic, they cannot outsource their compliance obligations. "Regulators will still expect banks to maintain their own risk-based AML controls, monitoring, escalation and oversight," he said. Banks should treat crypto as a higher-risk product, not just another trading feature, Shipkevich said. They should "vet vendors carefully, train AML staff to understand blockchain analytics, document alert decisions, and maintain independent compliance governance," he said. ### 相關股票 - [GBTC.US](https://longbridge.com/zh-HK/quote/GBTC.US.md) - [LTCN.US](https://longbridge.com/zh-HK/quote/LTCN.US.md) - [BITF.US](https://longbridge.com/zh-HK/quote/BITF.US.md) - [GLXY.US](https://longbridge.com/zh-HK/quote/GLXY.US.md) - [COIN.US](https://longbridge.com/zh-HK/quote/COIN.US.md) - [BTBT.US](https://longbridge.com/zh-HK/quote/BTBT.US.md) - [HUT.US](https://longbridge.com/zh-HK/quote/HUT.US.md) - [MARA.US](https://longbridge.com/zh-HK/quote/MARA.US.md) - [RIOT.US](https://longbridge.com/zh-HK/quote/RIOT.US.md) - [ETHE.US](https://longbridge.com/zh-HK/quote/ETHE.US.md) - [RIOX.US](https://longbridge.com/zh-HK/quote/RIOX.US.md) - [FBTC.US](https://longbridge.com/zh-HK/quote/FBTC.US.md) - [BRRR.US](https://longbridge.com/zh-HK/quote/BRRR.US.md) - [BSOL.US](https://longbridge.com/zh-HK/quote/BSOL.US.md) - [EETH.US](https://longbridge.com/zh-HK/quote/EETH.US.md) - [CONL.US](https://longbridge.com/zh-HK/quote/CONL.US.md) - [ETHT.US](https://longbridge.com/zh-HK/quote/ETHT.US.md) - [BLOK.US](https://longbridge.com/zh-HK/quote/BLOK.US.md) - [BTCO.US](https://longbridge.com/zh-HK/quote/BTCO.US.md) - [IBIT.US](https://longbridge.com/zh-HK/quote/IBIT.US.md) - [BITB.US](https://longbridge.com/zh-HK/quote/BITB.US.md) - [ETHV.US](https://longbridge.com/zh-HK/quote/ETHV.US.md) - [BITO.US](https://longbridge.com/zh-HK/quote/BITO.US.md) - [IMRA.US](https://longbridge.com/zh-HK/quote/IMRA.US.md) - [MSTX.US](https://longbridge.com/zh-HK/quote/MSTX.US.md) - [LMBO.US](https://longbridge.com/zh-HK/quote/LMBO.US.md) - [MRAL.US](https://longbridge.com/zh-HK/quote/MRAL.US.md) - [ETHA.US](https://longbridge.com/zh-HK/quote/ETHA.US.md) ## 相關資訊與研究 - [NEAR 協議 - 人工智能個人身份信息匿名化 - 日期（2026年5月20日）](https://longbridge.com/zh-HK/news/287125820.md) - [Bee Network celebrates its sixth "Bee's Day": opening 100,000 KYC slots and betting on AI games and prediction markets.](https://longbridge.com/zh-HK/news/287029478.md) - [Fireblocks 押注人工智能代理將推動下一輪穩定幣的繁榮](https://longbridge.com/zh-HK/news/287180724.md) - [中泰證券：Token 產業進入規模化落地階段 投資上重點關注三大方向](https://longbridge.com/zh-HK/news/287027652.md) - [Bella 協議與 NanoVita 達成合作夥伴關係](https://longbridge.com/zh-HK/news/286970691.md)