Scammers make crazy phone calls, AI calls back and forth, 2.5 million conversations, and 250,000 fraudulent intelligence reports.
How long has it been since you've received a phone call from a stranger?
While courier phones may also be intercepted at the same time, there is no way that there are too many phones for fraudsters and salespeople.
The headache is that crooks can always change their number, but we can't always not answer the phone.
Recently, an Australian AI company has come up with a new idea. If people don't want to take a scam phone call, let AI do it.
Moreover, the company's AI doesn't just answer phones, it actively pretends to be a victim, chats with scammers, and tries to get information out of each other's mouths.
Calls are picked up, emails are sent back and forth, shark links come in, and even investors are mixed up on social media platforms.
One of them said, "If you're trying to trick me, I'll play with you first".
Apate.AI, the company that sends the AI anti-fraud squad, was founded in 2024 and is based on cybersecurity research at Macquarie University in Sydney, Australia.
As of now, according to data released by the company, its AI-smart body has had 2.5 million conversations with scammers, cumulatively consuming the scammers about 17 years.
More importantly, these conversations also unearthed more than 250,000 fraudulent information.
Bank accounts, cryptocurrency wallets, malicious websites, phone numbers - all the clues that could have been found after the victims were deceived - are now coming straight out of the mouths of the scammers.
What's the business of letting AI talk to crooks?
What Apate.AI does is kind of like taking the honeycomb technology in the field of cybersecurity, moving it into the anti-fraud scene, and then using AI to scale it.
In the traditional field of cybersecurity, honeycombs are a specialized seduction system. For example, the defense deliberately builds a device that looks like a real business server and is actually used to monitor attack behavior.
The attacker believes he has found a target that can be captured, actively launches an attack, and the defense takes the opportunity to observe the tools, pathways and behavior patterns used by the other side, thus obtaining threat intelligence.
Instead of being tracked after an attack, it is better to set up temptations in advance to allow the attacker to actively expose himself.
Apate.AI's thinking is similar, except it's not about hackers, it's about scammers calling, texting, emailing, or searching for targets on social media platforms.
It will allow AI to disguise itself as a potential victim and actively enter the scenes of scammers' interactions.
Fraud calls are intercepted by AI; suspicious messages encountered, AI can respond; detecting potential fraudulent activity, text intelligence can also search forums, investment groups and social platforms and engage in a conversation as a virtual entity.
In order to make the disguise more realistic, these AIs can also simulate different ages, genders, accents and languages, making communication look as much as possible as communication between real users.
The scammer thinks he is targeting a potential victim, and in fact the other phone may be an AI specialized for intelligence.
Moreover, the AI victim not only doesn't easily hang up the phone, but can also ask scammers questions on purpose, according to a set goal.
What do you get out of talking to a crook for half an hour?
Apate.AI wants to extract critical information about fraudulent activities from these seemingly ordinary conversations.
For example, when fraudsters pretend to be bank customers, AI can try to access the phishing addresses, contact numbers and receivables they use. If they encounter cryptocurrency investment scams, smart entities can try to access the relevant wallet addresses.
In addition, the scams, fake brands, related websites, and other clues that could be used to track the gangs are all within the scope of intelligence collection.
According to TechCrunch, customers of Apate.AI can even directly assign tasks to a smartphone, such as asking it to try to access a bank account number or cryptocurrency wallet address.
This means that AI does not chat aimlessly, but interacts around a specific goal. At the end of the conversation, the system will also organize the information it obtains into intelligence for further analysis and use by customers.
Apate.AI also provides customers with intelligence and analytics platforms to help banks, telecom operators, and other agencies learn about fraud against their users, employees, brands, and products.
The whole process can be summarized in three steps:
Actively interacting:AI pretends to be a potential victim and engages in dialogue with fraudsters to get as much critical clues as possible.
It's the first time I've ever heard of it.The system organizes conversations, identifying phone numbers, accounts, wallet addresses, and malicious websites.
Assisted disposal:Customers use relevant information to identify, investigate and intervene on risks.
Compared to traditional anti-fraud processes, the biggest difference is that it attempts to get information ahead of time.
In the past, agencies often had to wait for user complaints, abnormal transactions, or fraudulent information to be reported to begin tracking. Apate.AI hopes to get investigative clues from interactions while fraud is still ongoing.
Of course, getting information doesn't mean having a complete network of scams.
Scammers may provide fake accounts, use one-time web addresses, or even intentionally spread misleading information. Therefore, clues obtained by AI still need to be verified, linked, and further disposed of by banks, telecom operators, or law enforcement agencies.
But if reliable information can be obtained in advance, agencies have the opportunity to act before more victims appear.
It's not just intelligence, it's also a waste of time for scammers.
It's not uncommon for a scammer to waste time on a phone. In the past, many people who specialized in chatting with a scammer purposely pretended not to know the operation, made others repeatedly explain, and even thought about how to delay a phone call for decades.
Apate.AI founder and CEO Dali Kaafar's initial entrepreneurial inspiration came from a similar experience.
A few years ago, he received a scam phone call while attending a family picnic. After realizing the identity of each other, he did not hang up directly, but decided to pretend not to understand each other's intentions and continue to chat with each other.
In the end, the call lasted 44 minutes.
This experience made Kaafar realize a question: If a normal person can drag a liar for 44 minutes, can it be automated?
One person can only answer one phone at a time, but thousands of AI entities have the opportunity to deal with a large number of fraud interactions at the same time.
This is what Apate.AI is trying to scale up. From an economic point of view, fraud is also an activity that seeks to put in and out of proportion. Scammers need to find targets, build trust, implement inducements, and finally figure out how to get victims to transfer money.
If a large amount of time is spent on an unsuccessful goal, it will theoretically increase the cost of the case.
Assuming a scammer has 8 hours a day to communicate, temporarily without considering additional time such as dialing, waiting and screening targets:
If the average interaction takes 3 minutes and 8 hours, theoretically 160 interactions can be completed.
If the average interaction takes 30 minutes and 8 hours, theoretically only 16 interactions can be completed.
The difference between the two is 10 times.
This is a simplified model of assumptions that does not represent the actual effectiveness of actual fraudulent activity.If AI is able to hold back fraudsters in a sustained, low-cost manner, it could increase the opportunity cost of each other reaching the real victims.
This is why Apate.AI would emphasize that its own AI intelligence body has cumulatively consumed about 17 years of fraudsters.
However, mere procrastination is not a complete anti-fraud scheme. Ideally, an interaction would produce two results simultaneously: both a waste of time for fraudsters and the ability to leave information available for subsequent investigations.
The former increases the cost of cases and the latter helps institutions to track and intervene. The overlap between the two is where the model really deserves attention.
Banks, telecoms operators, have started paying
As far as progress is concerned, Apate.AI is no longer satisfied with technical verification in the laboratory, but is entering the real business scene.
Clients disclosed by the company include Commonwealth Bank of Australia, telecom operator TPG Telecom, and other utilities, telecommunications companies and government agencies.
According to co-founder and chief commercial officer Brad Joffe, who told TechCrunch that TPG Telecom had transferred nearly 590,000 phone calls to Apate.AI's smartphone. Fraud losses reported by the Federal Bank of Australia decreased by 76%.
The data from the company's disclosures cannot simply be understood as AI alone has brought about all the improvements, but at least shows some of the downsides to this type of product.
For telecom operators, AI can take on suspicious incoming calls that are transferred, interact with each other and collect leads.
For banks, relevant information has the potential to be used to identify fraudulent activity, investigate linked accounts, and provide a basis for subsequent risk disposal.
For the government, the continuous accumulation of fraudulent intelligence can also serve as a reference for investigating related criminal activities.
So what Apate.AI really wants to sell is not a chatbot, but an anti-fraud system that revolves around AI interaction, intelligence gathering and subsequent disposal.
Customers buy not just the number of times they talk, but the information they get from these interactions, and the ability to turn that information into actual action.
AI is changing the effectiveness of defense on both sides
With the advent of Apate.AI, there is a larger industry background: AI is simultaneously reducing the cost of attack and defense.
For fraudsters, big models can help generate more natural phrases, and deep counterfeiting techniques can be used to mimic the voices and images of celebrities, corporate executives, or other credible figures.
The FBI Internet Crime Complaints Center (FBI IC3)'s 2025 annual report shows that more than 220,000 AI-related criminal complaints were filed in 2025, resulting in losses of about $893 million.
Fraudsters face increasing pressure when they start using AI to generate content, mass contact targets, and anti-fraud systems that rely solely on artificial identification and tracking.
Since attackers can be automated, the defense needs to find tools that can scale. Apate.AI's choice is to let AI directly participate in the interaction link in the fraud chain.
According to TechCrunch, the company uses both cutting-edge large models and fine-tuned proprietary models to support intelligence and intelligence analysis. Model training data includes historical conversations, as well as previous research by the founding team at Macquarie University.
As the scale of the interaction expands, the company hopes to continue accumulating fraudulent tactics, patterns of cases and related clues, and to use these data to improve its intelligence analytics capabilities.
However, this path is not without its challenges.
First, increasing the amount of data does not necessarily mean improving the quality of intelligence. How to filter out false information, identify false leads, associate different fraudulent activities, and avoid misjudgments will affect the actual value of such platforms.
Second, the success of AI in obtaining intelligence does not mean that fraud can be prevented in a timely manner. Whether accounts can be frozen, numbers can be intercepted, funds can be withdrawn still depends on the responsiveness of customer institutions and the collaboration between banks, telecom operators and law enforcement.
In other words, AI can be an important part of the anti-fraud system, but it cannot replace the whole anti-fraud system.
With more than $8 million in seed funding, Apate.AI is poised to continue its expansion, with the U.S. and Europe in its next stages.
Do you know how far China's AI anti-fraud has gone?
© THE