AI agents can be scammed
The Australian startup Apate.ai, a finalist at TechCrunch Disrupt and fresh off an $8.15 million seed round, deploys fleets of AI agents posing as scam victims: they answer scam calls, click phishing links, and hand over bank accounts and crypto wallets to the fraudsters. The intelligence is then sold on to banks and telecoms — though the impressive results are so far the company's own.
Fighting fraud has usually followed a sad sequence: first a real customer is tricked, then the bank learns which method the scammers used. The Australian startup Apate.ai reverses that order. The company sends thousands of AI agents out as bait — bots that allow themselves to be scammed, gather the evidence along the way, and sell the intelligence on to customers before real victims are hit.
The company presented itself as a finalist in the Startup Battlefield 200 at TechCrunch Disrupt in San Francisco, when TechCrunch profiled it on October 14, 2026 (TechCrunch). This came shortly after an $8.15 million seed round in August, with investors including Lobby Capital, OIF Ventures, Investible, Concept Ventures and Baobab Ventures — a round also confirmed by Whalesbook (Whalesbook). According to an executive at the company, referred to as Joffe in the TechCrunch article, the company has now raised just under $10 million in total, has 20 employees, and plans to spend the capital on product development and international growth — including an office in London and a larger presence in the US.
How the agent fleet works
Apate.ai was founded in 2024 on the basis of research at Macquarie University in Sydney. Technically, the platform consists of thousands of autonomous agents that pose as innocent victims. The agents can answer scam calls, respond to suspicious messages and emails, click phishing links, and join social media groups to lure in scammers, TechCrunch writes.
The agents vary in age, gender, accent and language, so that they appear to be a broad cross-section of potential victims. During the conversation, they extract what scammers would otherwise hide: money laundering bank accounts, crypto wallets, phone numbers, malicious websites, and which brands the scam campaign is impersonating. This information — which banks and telecoms normally only get after an actual victim has lost money — is fed into an insights portal where customers use it to tailor their anti-fraud measures.
The engines behind it are a combination of so-called frontier models and the company's own fine-tuned models, Joffe tells TechCrunch. The training data consists partly of conversations the agents have already had with scammers, partly of the Macquarie research led by the researcher Kaafar — a direct link between the basic research and the commercial product. Customers can also assign the agents specific tasks from the portal, such as "get a bank account" or obtain a crypto wallet address for an active scam campaign.
The numbers — and who stands behind them
The company's own figures, put forward via Joffe in the TechCrunch interview, are extensive: the agent fleet is said to have so far wasted 17 years of scammers' time across 2.5 million conversations. TPG Telecom is said to have redirected almost 590,000 calls to Apate's agents. Commonwealth Bank of Australia is said to have seen a 76 percent drop in publicly reported scam losses. And an unnamed Australian bank is said to have used the agents actively — with tasks including finding a bank account — to disrupt 22 scam campaigns.
It is worth emphasizing: none of these numbers have been independently verified. They come from the company itself, relayed by TechCrunch, and Whalesbook also repeats them as the company's own results. It is unclear whether TPG or Commonwealth Bank have themselves publicly confirmed the metrics. The $8.15 million round, by contrast, has been confirmed by a source other than TechCrunch.
The timing is no coincidence
Interest in Apate coincides with authorities beginning to measure AI-driven fraud directly. For the first time in the more than twenty-year history of IC3, the FBI's Internet Crime Report is tracking AI-related complaints (WBRC). In 2025, more than 22,000 such complaints were filed, with around $893 million in losses — of which over $632 million came from investment fraud alone, where the FBI says scammers use AI to conduct conversations and create deepfakes of celebrities, executives and other trusted figures.
The picture is, moreover, two-sided: AI is not only used by defenders. On October 8, 2026, OpenAI said it had shut down two covert influence campaigns, one from Russia and one from Iran, that used ChatGPT to build up what the company called "false front" entities — and assessed the Russian campaign as the first category 5 influence operation OpenAI has disrupted (Unite.AI). Apate's business idea thus presupposes a world in which scammers themselves get better at using AI — and in which the defense must be industrialized to keep up.
Open questions
The model also has unresolved aspects. Whalesbook points out that the systematic use of false identities, even when directed at criminals, may fall into a legal gray zone under privacy and telecoms legislation in various countries — an analysis, not an established fact, and no regulator has commented on the matter in the source material.
There is also a technical counterargument: if scammers understand that a share of their "victims" are bots, they can train their own systems to recognize bot-driven conversations and filter them out — a form of arms race that Whalesbook points to as a possible consequence. And most fundamentally: as long as the headline numbers — the 17 years, the 2.5 million conversations, the 76 percent drop and the 590,000 calls — come only from the company itself, the entire value proposition rests on Apate's own accounting of its own effect. Independent verification of the customer metrics is so far lacking.