Apate 派遣 AI Agent 卧底受骗，收集诈骗情报

大多数公司利用 AI Agent 更快地交付产品或加速工作流程。Apate.ai 则派它们卧底，让它们去“受骗”。

这家澳大利亚创业公司成立于 2024 年，其基础来自悉尼麦考瑞大学开展的研究。公司部署了数千个伪装成无辜受害者的 AI Agent。这些 Agent 可以接听诈骗电话、回复可疑短信和电子邮件、访问钓鱼链接，并加入社交媒体群组诱骗诈骗分子。它们会假扮成各种类型的人，拥有不同的年龄、性别和口音，并使用多种语言，努力让诈骗者尽可能长时间地保持互动，从中获取有用信息。

该公司表示，其 Agent 集群目前已在 250 万次对话中累计浪费诈骗分子 17 年时间。更重要的是，对于 Apate 的客户群——包括银行、电信公司、政府机构和公用事业企业——这些卧底 Agent 会尝试让诈骗者透露通常只有在受害者已经被骗 之后 才能获知的信息：银行账户、加密货币钱包、电话号码、恶意网站、其冒充的品牌，以及与其诈骗活动有关的其他细节。

Apate 是今年 TechCrunch Disrupt Startup Battlefield 200 的入围者之一，随后会将这些信息提供给客户，客户可利用这些信息制定反欺诈措施和行动。（该公司将在 Disrupt 向由风投组成的评审团进行路演。活动将于本周在旧金山 Moscone West 举行，持续至 10 月 15 日星期四。）

这是诈骗诱捕手段的一种有趣演变，但考虑到 AI 已在多大程度上赋能诈骗者和黑客，这并不令人意外。FBI 在 2025 年收到了超过 2.2 万起与 AI 相关犯罪的投诉，造成约 8.93 亿美元损失。其中超过 6.32 亿美元来自投资诈骗；FBI 表示，诈骗者会利用 AI 运营对话，并深度伪造名人、高管和受信任人士的声音与形象。

长期以来，人们一直将故意拖住诈骗者当作一种消遣——有时效果颇为滑稽，例如让诈骗者朗读整本书，甚至让他们去纹荒谬的纹身。如今，随着 AI 能力持续增强，局面正在反转。

有意思的是，Apate 背后的想法源自一通真实的诈骗电话。

几年前，Apate 创始人兼 CEO Dali Kaafar 正在参加家庭野餐，这时接到一个陌生号码来电。意识到电话另一端是诈骗者后，他决定为了孩子们的乐趣装傻，并成功让通话持续了 44 分钟。

所有人都笑得很开心，但 Kaafar 注意到，这 44 分钟里诈骗者没有去拨打真正受害者的电话。这也是他自己被浪费掉的 44 分钟，于是他开始思考：能否将这件事自动化，并在这个过程中学到一些有价值的信息？

这一想法最终促成了他在麦考瑞大学任职期间开展的研究工作。Kaafar 与同事利用机器学习和自然语言处理技术，识别诈骗电话的内容，并开发更好的诈骗检测与防范方法。

“行业中的大多数工作都发生在事后。电信运营商会封禁号码，但诈骗者只需换一个新号码。银行会监控交易，但到那时诈骗者已经接触过客户。我们直接去找诈骗者，在任何人受害之前获取信息，”联合创始人兼首席商务官 Brad Joffe 告诉 TechCrunch。“我们的 AI Agent 每天通过电话、消息、电子邮件和网络与数万名诈骗者交谈。我们向客户提供的是这些对话中产生的信息，以及客户能够如何利用这些信息。”

Joffe 表示，Apate 同时使用前沿模型和自主微调模型来驱动其 Agent 和分析系统。公司利用 Agent 与诈骗者进行的对话，以及 Kaafar 在麦考瑞大学开展的研究来训练模型。

除能够使用 Apate 的 Agent 集群外，客户还可以访问一个洞察门户，该门户提供关于针对其终端用户或员工、品牌、产品等对象的诈骗活动的情报与分析。客户甚至可以给这些 Agent 分配特定任务，例如获取一个银行账号或加密货币钱包地址。其文本 Agent 可以在论坛、投资群组和社交平台中搜索潜在诈骗活动，随后以虚构身份进入对话，挖掘诈骗者的详细信息。

这家创业公司的客户包括澳大利亚联邦银行、TPG Telecom，以及其他公用事业企业、电信运营商和政府机构。Joffe 表示，TPG 已将近 59 万通电话转接至 Apate 的 Agent；澳大利亚联邦银行公开报告的诈骗损失下降了 76%；而一家未具名、与该创业公司合作的澳大利亚银行，则利用其外呼 Agent 干扰了 22 个诈骗活动。

投资者也押注于“愚弄诈骗者”能够带来商业价值。Apate 在 8 月从 Lobby Capital、OIF Ventures、Investible、Concept Ventures 和 Baobab Ventures 获得了 815 万美元种子轮融资。Joffe 表示，这家拥有 20 名员工的公司累计融资额略低于 1000 万美元，计划将最新一轮资金用于产品开发和国际扩张，包括在伦敦设立办公室，并扩大其在美国的业务布局。

“Apate 以希腊神话中的欺骗女神命名，考虑到我们的 Agent 所做的事情，这个名字很合适，”Joffe 说。他补充称，这家创业公司甚至会在路演材料中展示诈骗者辱骂机器人的内容。

“我们将其视为挫败感的信号，Apate 的每一句脏话都会被庆祝。我们可能是全球唯一一家会因为人们辱骂产品而庆祝的企业。”

Most companies use AI agents to ship products faster or to speed up workflows. Apate.ai sends them undercover to get scammed.

Founded in 2024 on the basis of research done in Macquarie University in Sydney, the Australian startup deploys thousands of AI agents that masquerade as innocent victims. The agents can answer scam calls, respond to shady messages and emails, follow phishing links, and join social media groups to bait fraudsters. They pretend to be people of all stripes, sporting different ages, genders, and accents, and speak various languages as they try to keep scammers engaged long enough to learn something useful.

The company says its fleet of agents has so far wasted 17 years of scammers’ time across 2.5 million conversations. More importantly for Apate’s customer base of banks, telecom companies, government agencies, and utilities, the undercover agents try to get scammers to reveal things about themselves that are usually learned only after a victim has been scammed: bank accounts, crypto wallets, phone numbers, malicious websites, the brands they’re impersonating, and other details about the scams they’re running.

Apate, one of the finalists at this year’s Startup Battlefield 200 at TechCrunch Disrupt, then feeds that information to its customers, who use it to inform their anti-fraud measures and initiatives. (The company will pitch a panel of VC judges at Disrupt, which is taking place this week at Moscone West in San Francisco through Thursday, October 15.)

It’s a curious evolution of scam-baiting, but it’s not so surprising given how much AI has empowered scammers and hackers. The FBI received more than 22,000 complaints of AI-related crimes in 2025, which resulted in about $893 million in losses. More than $632 million of that came from investment scams, in which the FBI says fraudsters use AI to run conversations, and deepfake the voices and appearances of celebrities, executives, and trusted figures.

For a long time, people have made a hobby out of deliberately stringing scammers along — sometimes hilariously, like getting scammers to read aloud entire books, or even to get ridiculous tattoos. Now, with AI growing increasingly capable, the tables are turning.

Curiously, the idea behind Apate was sparked by a real scam call.

Several years ago, Apate’s founder and CEO Dali Kaafar was at a family picnic when he got a call from an unknown number. Realizing it was a scammer on the line, he decided to play dumb for the benefit of his kids and managed to keep the call going for 44 minutes.

Everyone had a laugh, but Kaafar noticed that those were 44 minutes the scammer had not spent calling a real victim. It was also 44 minutes of his wasted time, which led him to wonder: Could he automate it, and in the process, learn something useful?

That idea eventually led to research work at his job at Macquarie University, where Kaafar and his colleagues used machine learning and natural language processing to uncover the content of scam calls, and develop better ways to detect and prevent scams.

“Most of the industry works after the fact. Telcos block numbers, but scammers just get new ones. Banks watch transactions, but by then the scammer has already reached the customer. We go to the scammer directly and get the information before anyone gets hurt,” Brad Joffe, co-founder and chief commercial officer, told TechCrunch. “Our AI agents talk to tens of thousands of scammers every day across calls, messages, email, and the web. What we offer our customers is what comes out of those conversations, and what they can do with it.”

Joffe said Apate uses both frontier models and its own fine-tuned models to power its agents and analysis. The company trains its models on conversations its agents have had with scammers, as well as the research Kaafar did at Macquarie.

In addition to access to Apate’s fleet of agents, customers get access to an insights portal that offers intelligence and analytics on scams targeting their end users or employees, brands, products, and more. Customers can even assign these agents particular missions, such as obtaining a bank account number or crypto-wallet address. Its text agents can search forums, investment groups, and social platforms for potential scam activity, and can then enter conversations with a fabricated identity to surface scammers’ details.

The startup’s customers include Commonwealth Bank of Australia, TPG Telecom, as well as other utilities, telcos, and government agencies. Joffe said TPG has diverted nearly 590,000 calls to Apate’s agents, Commonwealth has seen a 76% drop in publicly reported scam losses, and an unnamed Australian bank the startup works with has used its outbound agents to disrupt 22 scam campaigns.

Investors are betting there is money to be made from fooling scammers, too. Apate raised $8.15 million in seed funding in August from Lobby Capital, OIF Ventures, Investible, Concept Ventures, and Baobab Ventures. Joffe said the 20-person company has raised just under $10 million in total and plans to spend the latest round on its product and international expansion, including opening an office in London and building a larger presence in the U.S.

“Apate is named after the Greek goddess of deception, which felt appropriate given what our agents do,” Joffe said, adding that the startup even features scammers swearing at the bots in its pitch material.

“We track that as a sign of frustration and every swear word gets celebrated by us at Apate. We are probably one of the only businesses in the world that celebrates when people swear at our product.”