Scammers often pretend to be someone else to get money or personal information, but they can be fooled themselves: some artificial intelligence systems can hold conversations by making the interlocutor believe they are talking to a real person. A rather curious use for a technology that can also be used to clone voices and make telephone scams more credible.
An example comes from Australia, where the startup Apate claims that its bots have already managed millions of conversations. The goal is not only to keep bad guys busy, but above all to collect information useful for identifying fraud.
How AI pretends to be a scam victim
Pretending to be interested in a scammer’s requests is a strategy that already existed before the arrival of artificial intelligence. She is known as scam-baiting and is practiced by people who voluntarily respond to suspicious phone calls or messages, with the intention of prolonging the conversation without falling into deception.
Today this activity can also be carried out by AI agents, capable of autonomously holding a conversation and adapting their responses to those of the interlocutor. Rather than following a pre-established script, they can ask questions, pretend not to understand and change the subject, maintaining a consistent personality.
In phone calls the process can occur in three steps. The system transcribes the words of the caller, a linguistic model processes the response and a voice synthesizer transforms it into a voice that the interlocutor hears almost in real time. The result is a dialogue in which the scammer may not realize he is talking to a machine. In chats, it is enough to process the written answers.
To receive calls, some systems leverage collaboration with telephone operators, who can direct suspicious contacts to AI agents. Other projects instead use “bait” numbers, specifically set up to attract fraudulent calls.
The Australian case: How artificial intelligence gathers fraud intelligence
One of the most complex applications of this approach is that of Apate, an Australian startup born in 2024 from a research project at Macquarie University in Sydney. During a family lunch, researcher Dali Kaafar received a fraudulent phone call and decided to entertain the caller for about 40 minutes. He thus managed to keep him busy, preventing him from looking for other possible victims, but he also lost time. That’s when he thought about automating this task.
The research team analyzed telephone recordings and message exchanges to study the scammers’ techniques and developed bots with different identities, ages, voices and accents, capable of conversing in multiple languages.
Their use is not limited to telephone calls. Apate agents can also respond to SMS and emails, interact on platforms such as WhatsApp and Telegram and join online groups in which, for example, suspicious investments are promoted.
To carry out a scam, the person organizing it could communicate the details of a bank account, the address of a cryptocurrency wallet, or a link to a fraudulent site. The bots try to surface these details, along with the phone numbers used and the names of the companies imitated. The collected data is then made available to customers to identify new campaigns and improve security measures.
According to what the startup released in October 2026, its agents have already handled 2.5 million conversations, keeping scammers busy for the equivalent of 17 years. This last data represents the sum of the duration of the individual exchanges. The company also claims to have collected over 250,000 pieces of information on fraudulent activities.
In its report published in March 2026, Australia’s National Anti-Scam Center reports that Commonwealth Bank uses thousands of bots to gain insights into fraud and strengthen detection systems. The document also describes a trial by TPG Telecom, during which bots kept scammers busy for the equivalent of seven months in total. Apate’s technology is primarily intended for organizations, not individual users.
How effective are anti-scam bots and what are their limits?
The Australian one is not the only example of the use of artificial intelligence against scammers. In November 2024, the British operator O2 introduced Daisy, a virtual “grandmother” capable of entertaining interlocutors by telling stories about her family, talking about knitting and rambling when someone asked her for money or personal information. According to the company, Daisy was able to extend some calls to around 40 minutes. In a subsequent update, O2 reported that the bot had responded to over 1,000 calls, also allowing it to identify some of the strategies used by the attackers.
The numbers communicated by the companies help to understand the scope of these initiatives, but require clarification. Two and a half million conversations do not correspond to as many foiled scams, just as the time wasted by criminals does not allow us to automatically calculate how much money has been saved. To evaluate the real effectiveness, independent data on avoided fraud and reduced economic losses would be needed.
An attacker can notice the deception, interrupt the contact or use multiple numbers and accounts. These systems only reach fraud attempts that are detected and directed towards the agents, or those intercepted in the online spaces in which they operate.
For this reason, bots represent a complementary tool to other forms of protection, such as filters that block calls with falsified numbers, introduced in Italy by AGCOM.
However, the attempt to entertain scammers should not be imitated when you receive a suspicious phone call. Doing so voluntarily may expose you to further manipulation or collection of personal data. It is therefore more prudent to stop communication and verify any requests through official channels.
