What is new

Apate.ai, an Australian startup founded in 2024 out of research at Macquarie University in Sydney, has built a fleet of AI agents that do the opposite of what most customer-facing bots do. Instead of helping real users, they pretend to be potential scam victims. The agents answer scam phone calls, reply to suspicious texts and emails, click through phishing links, and even join social media groups and investment forums where fraudsters recruit targets. Their goal is not to avoid being scammed but to stay on the hook as long as possible, drawing out details that normally only surface after a real person has already lost money.

The company, a finalist in this year's Startup Battlefield 200 at TechCrunch Disrupt, says its agents have so far held 2.5 million conversations with scammers and collectively burned through 17 years of fraudsters' time. For its customers, banks, telecom carriers, utilities and government agencies, the real value is not the wasted time itself but the intelligence that falls out of these conversations: bank account numbers, cryptocurrency wallet addresses, phone numbers, malicious websites, and the brands or institutions a scam is impersonating.

How it works

Apate's agents are built to sound like plausible targets rather than obvious bait. They are designed to vary in apparent age, gender, accent and language, and they can be given specific missions by a customer, such as trying to extract a bank account number or a crypto wallet address from a particular scam campaign. Some agents work reactively, picking up calls or messages that are routed to them. Others work proactively, searching forums, investment groups and social platforms for signs of scam activity and then striking up a conversation under a fabricated identity to draw the scammer out.

Behind the scenes, the company says it combines frontier large language models with its own fine-tuned models, trained on the corpus of real scam conversations its agents have collected, as well as on the earlier academic research into scam-call detection that its founder, Dali Kaafar, carried out at Macquarie University using machine learning and natural language processing. In plain terms, a large language model is a system trained on huge amounts of text to predict and generate plausible, context-appropriate replies; fine-tuning means further training that general model on a narrower set of examples, in this case thousands of logged scam interactions, so it responds in ways tailored to that specific task. Customers access the results through what Apate calls an insights portal, which surfaces analytics on the scams aimed at their own customers, employees, brands or products.

Why it matters

The pitch behind Apate is that most anti-fraud work today happens after damage is already done. Telecom companies block numbers once they are reported, but scammers simply switch to new ones. Banks monitor transactions for suspicious patterns, but by the time a transaction looks wrong, the scammer has usually already reached the victim. Apate's co-founder and chief commercial officer, Brad Joffe, frames the company's approach as going directly to the scammer to gather information before any real customer is harmed, rather than reacting once harm has occurred.

The backdrop is a rapid rise in AI-enabled fraud. According to the FBI, it received more than 22,000 complaints of AI-related crimes in 2025, tied to roughly $893 million in losses, with more than $632 million of that coming from investment scams in which fraudsters reportedly use AI to run conversations and to deepfake the voices and appearances of celebrities, executives and other trusted figures. Against that trend, Apate's bet is that fighting AI-driven scams increasingly requires AI on the defensive side too.

Apate lists Commonwealth Bank of Australia and TPG Telecom among its customers, alongside other utilities, telcos and government agencies. Joffe cited specific figures from this customer base: TPG has diverted nearly 590,000 calls to Apate's agents, Commonwealth Bank has seen a 76 percent drop in publicly reported scam losses, and an unnamed Australian bank has used Apate's outbound agents to disrupt 22 scam campaigns. The announcement does not explain the methodology behind these figures or how the drop in reported losses was measured, so it is worth treating them as company-reported outcomes rather than independently verified results.

  • Founded 2024 in Sydney, based on research at Macquarie University by CEO Dali Kaafar
  • Agents have held 2.5 million conversations with scammers, consuming 17 years of scammer time, per the company
  • Customers include Commonwealth Bank of Australia and TPG Telecom, among other telcos, utilities and government bodies
  • Reported customer results: TPG diverted nearly 590,000 calls to Apate agents; Commonwealth Bank saw a 76% drop in publicly reported scam losses; one bank disrupted 22 scam campaigns
  • Raised $8.15 million in seed funding in August 2026 from Lobby Capital, OIF Ventures, Investible, Concept Ventures and Baobab Ventures, bringing total funding to just under $10 million
  • 20-person company, planning offices in London and a larger U.S. presence

Limits and open questions

The source material does not detail how Apate verifies the accuracy of the information scammers reveal during these staged conversations, nor how quickly that intelligence is turned into concrete blocks or warnings for end customers. It also does not specify what fraction of scam campaigns targeted at a given bank or telco are actually intercepted by Apate's agents versus reaching real people, which would be useful context for the reported drop in losses at Commonwealth Bank. There is no detail on false positives, that is, whether the system might waste effort on genuine callers mistakenly flagged as scammers, or on how the company handles the ethical and legal questions raised by agents adopting fabricated identities, including impersonating specific demographic groups, across different jurisdictions.

It is also not clear from the announcement how Apate measures success internally beyond the time-wasted and conversation-count metrics it publicizes, or how its fine-tuned models are evaluated against newer, more sophisticated scam tactics as fraudsters adapt. The company's own framing, that it tracks scammers swearing at its bots as a marker of success, suggests an informal rather than a rigorously benchmarked evaluation approach, at least based on what has been disclosed publicly.

How to try it

Apate is a business-to-business product rather than a consumer app. The announcement does not describe a self-serve signup process; instead, the company works directly with banks, telecom carriers, utilities and government agencies, who get access to the agent fleet along with the insights portal for scam analytics. Readers curious about the product would need to go through the company directly, as no public trial or demo access is mentioned in the source material.

Sources