Apate.ai Sends AI Agents Undercover to Get Scammed

Most companies deploy AI agents to get work done faster. Apate.ai deploys them to be fooled. The Australian startup runs thousands of agents that pose as ordinary people, pick up scam calls, answer suspicious messages and play along until the fraudster gives something away.

The company grew out of research at Macquarie University in Sydney and was founded in 2024. This week it is one of the finalists in the Startup Battlefield 200 at TechCrunch Disrupt, the startup conference running at Moscone West in San Francisco through Thursday, October 15, where it will pitch to a panel of venture capital judges.

How the decoys work

Apate's agents work across channels. They answer phone calls, reply to emails and text messages, click through phishing links and join social media groups where scammers look for targets. Each one plays a different persona, with its own age, gender, accent and language. The goal is to keep the scammer talking long enough to learn something.

By the company's count, its agents have so far taken up 17 years of scammers' time over 2.5 million conversations. Wasting a criminal's afternoon is only part of the pitch, though. What Apate sells is the information that comes out of those exchanges: bank account details, crypto wallet addresses, phone numbers, malicious websites, the brands being impersonated and how a given scam is set up. Normally, that kind of detail only surfaces after a real victim has lost money.

Customers - banks, telecom operators, utilities and government agencies - use the data to shape their own anti-fraud work. They also get an insights portal with analytics on scams aimed at their customers, staff, brands and products. They can hand agents specific missions, such as extracting a particular bank account number or wallet address. Text-based agents can search forums, investment groups and social platforms for suspected scams, then join the conversation under a made-up identity.

Co-founder and chief commercial officer Brad Joffe describes the approach as moving earlier in the chain. "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," he told TechCrunch. "We go to the scammer directly and get the information before anyone gets hurt."

According to Joffe, the agents run on a mix of frontier models and Apate's own fine-tuned models. Training data includes the company's past conversations with scammers and the earlier Macquarie research.

A 44-minute phone call

The idea began with a real scam call. Several years ago, founder and CEO Dali Kaafar was at a family picnic when an unknown number rang. He recognised a scammer on the line, played dumb to entertain his kids and kept the caller busy for 44 minutes.

Kaafar noticed two things. Those 44 minutes were time the scammer did not spend on a real victim. They were also 44 minutes of his own time. That raised the question of whether the job could be automated, and whether something useful could be learned along the way. At Macquarie, he and his colleagues went on to apply machine learning and natural language processing to the content of scam calls, with the aim of improving detection and prevention.

Scam-baiting itself is nothing new. Hobbyists have spent years stringing fraudsters along, sometimes getting them to read entire books aloud or even get absurd tattoos. Apate turns that pastime into a service.

Early numbers from customers

The startup's named customers include Commonwealth Bank of Australia and TPG Telecom. Joffe said TPG has routed nearly 590,000 calls to Apate's agents, and Commonwealth Bank has seen a 76% drop in publicly reported scam losses. An unnamed Australian bank has used Apate's outbound agents to disrupt 22 scam campaigns. These figures come from the company and have not been independently verified.

The backdrop is a sharp rise in AI-assisted fraud. The FBI received more than 22,000 complaints about AI-related crimes in 2025, tied to roughly $893 million in losses. More than $632 million of that came from investment scams, where the bureau says criminals use AI to run conversations and to fake the voices and faces of celebrities, executives and other trusted figures.

In August, Apate raised $8.15 million in seed funding from Lobby Capital, OIF Ventures, Investible, Concept Ventures and Baobab Ventures. The 20-person company has raised just under $10 million in total. It plans to spend the new money on product work and international growth, including a London office and a bigger presence in the U.S.

The name comes from Apate, the Greek goddess of deception. Joffe said the team even treats swearing as a metric: abusive scammers are a sign the bots are working. "We are probably one of the only businesses in the world that celebrates when people swear at our product," he said.

Our Take

Apate is a clear case of the same technology being used on both sides of a fight. Fraudsters already use AI to scale conversations and clone voices, and recent cases like the AI-assisted breach of Korean banks show how far one attacker can now get with the right tools. Apate's answer is to scale the decoys too, much as other startups are building AI models aimed at defenders.

The more interesting part is the data, not the wasted hours. If agents can reliably pull account numbers and wallet addresses before money moves, banks and telcos get a head start they rarely have now. Still, the customer figures are the company's own, and a drop in reported losses can have several causes.

It is worth watching whether scammers learn to spot AI personas, which could turn this into an arms race between bots. Expansion into the UK and U.S. will also test how well the model travels beyond Australia, where its early partners are based.