AI Is Becoming a New Weapon Against Scammers

Artificial intelligence is no longer just helping criminals automate fraud; it is also being turned against them. Security researchers and companies are now using AI systems to waste scammer time, gather intelligence, and disrupt the industrial-scale operations behind phone fraud, text scams, and online deception.

One of the clearest examples is Apate, an Australian company that has spent the last two years building a network of AI “victims” designed to keep scammers on the line for as long as possible while recording the details of their schemes. The company says banks and telecom partners are already using the system, which now includes hundreds of thousands of bots.

The rise of these tools matters because global authorities have struggled to keep up with cybercrime, particularly when attackers operate across borders and at massive scale. As scams become faster, more convincing, and more automated, defenders are increasingly trying a different strategy: don’t just block the scam — distract, delay, and document it.

Why defenders are turning AI on scammers

Online fraud has become too large and too distributed for traditional enforcement alone. Criminal groups can launch billions of calls, texts, and messages each year, often using automated dialers, spoofed identities, and rapidly shifting infrastructure to evade detection. That has pushed banks, telecom carriers, platform providers, and investigators to search for methods that do not rely solely on shutting down one offender at a time.

Artificial intelligence fits that mission because it can do two things at once: act like a believable target and collect useful intelligence. Instead of simply filtering a suspected scam attempt, an AI system can hold the attacker in conversation, draw out details about the operation, and potentially prevent the scammer from contacting a real victim during that same window of time.

This approach does not solve cybercrime on its own. But it offers defenders a way to turn one of scammers’ biggest strengths — scale — into a weakness. Every minute spent persuading a fake victim is a minute not spent reaching an actual one.

“A minute that a scammer is talking to a bot or an agent is a minute where you’re probably saving hundreds, if not thousands of possible people being reached out to by that exact same scammer,” Apate founder and chief executive Dali Kaafar said, describing the value of keeping fraudsters occupied.

What is Apate, and how does it work?

Apate is an Australian anti-scam platform named after the Greek goddess associated with deceit. The company builds AI personas that answer calls, reply to messages, and even enter scam chat groups online. The goal is simple: make fraudsters believe they have found a real target, then keep them engaged long enough to expose how the scam works.

Kaafar says the platform is used by banks and backed by telecom companies. He says it now includes about 350,000 bots, each designed to behave a little differently so the deception feels less robotic and more human. Some of the bots answer calls. Some respond in text chats. Others are configured to delay, deflect, or create just enough uncertainty to keep the fraudster interested.

The company says the system has already collected more than 250,000 pieces of live fraud intelligence, including scam links, bank-account details, and money-mule information. That kind of data can help institutions identify campaigns faster and build richer patterns around how criminals move money and contact victims.

How the bots avoid giving themselves away

The AI personas are built to seem inconsistent in realistic ways. According to Kaafar, they may have different language abilities, different levels of confidence, and different profiles. That variety is important because scammers are trained to detect scripted or too-perfect responses.

In practice, the bots may sound like people with partial attention spans, uncertain financial knowledge, or slightly messy personal habits. A scammer might get one personality that answers a call immediately and another that says it will call back later. Some may have messaging apps such as WhatsApp; others may not. The point is not to appear flawless. The point is to appear plausible.

Kaafar said the bots are intentionally given human-like quirks, including the kind of inconsistent availability that real people show, so scammers do not quickly realize they are being played.

The result is an anti-fraud system that does not merely detect suspicious behavior. It actively participates in the deception, but on behalf of the intended victim rather than the criminal.

Can an AI bot really waste a scammer’s time?

Yes, and the whole idea depends on that delay being convincing enough to keep the scammer invested. WIRED tested a demo version of Apate in which the user takes the role of the scammer and tries to persuade one of the AI victims. The exchange was designed to produce the same kind of friction a fraudster would face in a real call.

The test showed how the system can create the psychological effect defenders want. The bot was skeptical but not dismissive, leaving enough openings for a scammer to keep pushing. That balance matters because a scammer who realizes quickly that they are not dealing with a real person will simply hang up and move on.

During testing, the interaction felt natural enough to keep the conversation going. The researchers and editors behind the test were unable to get the system to accept a cryptocurrency pitch, even after several minutes of effort. That failure was the point: the AI should be frustrating, opportunistic, and just believable enough to keep the scammer from giving up.

Kaafar says some conversations can run for more than two hours. Even if many scam attempts end earlier, that long-tail value is important. Large-scale scam operations depend on volume and speed. Anything that slows the machine can reduce the number of real targets they reach in a day.

What the test revealed

• The AI responded quickly enough to feel conversational.
• The persona gave mixed signals rather than a hard refusal.
• The scam pitch did not collapse the interaction immediately.
• The setup suggested the system could absorb a lot of attacker attention.

That combination — responsiveness plus ambiguity — is what makes the defense approach useful. The bot does not need to “win” every conversation in the sense of exposing the scammer. It needs to occupy enough time, enough often, to disrupt the economics of fraud.

How big is the scam problem AI is trying to solve?

The scale is one of the main reasons these experiments are gaining traction. Cybercriminals now run highly organized operations that can resemble industrial workplaces more than loose groups of hackers. In some regions, scammers work out of physical facilities designed for high-volume fraud, using staff, scripts, and automated systems to contact as many people as possible.

That scale has made a purely reactive approach less effective. If a bank or police agency takes down one fraud ring, another can emerge elsewhere, often with slightly different infrastructure and the same basic tactics. The global, cross-border nature of the crime also limits what any one authority can do.

As a result, the defense strategy is shifting from simple blocking to active disruption. The aim is to waste the attacker’s resources, identify trends sooner, and make it more difficult for criminal networks to operate at full speed.

That table helps explain why this area is attracting investment and experimentation. Human scambaiters and researchers have done valuable work for years, but AI expands the volume and persistence defenders can bring to the task.

What are honeypots, and why are they changing?

Honeypots are fake systems set up to lure attackers. Security teams have used them for years to study intrusions, misdirect hackers, and collect data on malicious tools and methods. Traditionally, these traps have included decoy servers, virtual machines, or other digital environments that look valuable but are actually monitored.

The latest shift is the addition of large language models. Researchers say LLMs can make honeypots appear more lifelike by generating responses and system behavior that are less predictable than older scripted traps. That unpredictability can matter because experienced attackers often recognize static decoys.

Mark Vero, a doctoral researcher at ETH Zurich, says open-source honeypot providers are already working LLMs into their systems. The reason is straightforward: if the fake environment looks and behaves more like a real one, attackers may stay longer and reveal more.

Vero said in recent research that LLM-powered honeypots kept AI agents attacking the system for significantly longer than more predictable decoys, suggesting that realistic behavior can meaningfully improve defensive outcomes.

Vero and his colleagues also found that agentic attackers were less likely to identify the LLM-based honeypots as traps. In other words, the smarter the attacker becomes, the more important realism becomes on the defender’s side.

Why psychological manipulation is now part of cyber defense

For years, most security tools have focused on detection, blocking, and attribution. But as criminals become more adaptable, some defenders are thinking more like intelligence operators. They are trying to understand how to exploit the behavioral weaknesses of fraud networks rather than only their technical flaws.

That includes tactics designed to provoke curiosity, greed, impatience, or overconfidence. A scammer who believes a potential victim is close to paying may continue the call or message thread far longer than makes economic sense. If the AI can sustain that illusion, it can turn the criminal’s own persistence into a defensive asset.

There is also a broader intelligence benefit. Conversations can expose payment routes, scam scripts, mule-account details, URLs, and associated accounts. Those pieces of information can help banks, carriers, and law enforcement connect patterns that would otherwise remain fragmented.

In that sense, AI anti-scam tools are not only blocking a threat. They are creating a live feed of how the threat evolves in real time.

What does this mean for banks, telecoms, and platforms?

Banks, telecom operators, and online platforms are under pressure to do more because they sit at different points in the fraud chain. Banks see the money movement. Carriers see the phone traffic. Platforms may see scam communities and messaging channels. AI systems that can span those spaces may help join up information that is usually siloed.

The promise here is better collaboration. If a scam call leads to a suspicious account number, and that account number links to repeated cash-outs or mule activity, automated analysis can help surface the pattern sooner. If a text scam contains the same URL cluster as previous campaigns, AI can flag it before more users are hit.

That does not replace human judgment. But it can speed up triage and make it easier for investigators to prioritize the most dangerous operations.

Potential benefits for defenders

1. More time wasted for scammers.
2. More fraud intelligence collected at scale.
3. Better pattern detection across institutions.
4. Stronger disruption of repeat campaigns.

Each of those benefits becomes more powerful when combined. A bot that wastes a scammer’s time and also feeds intelligence systems can do the work of a small team at a much larger scale.

What are the risks of fighting scams with AI?

As with any automated security tool, effectiveness depends on careful design. If the bot sounds too artificial, it will be dismissed. If it is too persuasive, there are ethical and legal questions about how much deception defenders should use. If it scales too broadly without oversight, it could create false positives or capture data that must be handled carefully.

There is also the issue of escalation. Criminals are already using AI to write more convincing messages, generate voices, and personalize attacks. That means the defensive side may need to keep improving its realism just to stay in the same place.

Still, the trajectory is clear. Defenders are increasingly using the same class of tools that enabled more efficient fraud to build more efficient resistance. The arms race is not only technical anymore. It is behavioral.

How AI may change anti-scam work next

AI is unlikely to end online fraud by itself, but it could reshape how the fight is organized. The most effective future systems will probably combine automated bots, human investigators, threat sharing, telecom filtering, financial monitoring, and platform enforcement. The goal is to create a networked response rather than isolated reactions.

That may also push security teams toward more proactive tactics. Instead of only waiting for a complaint, institutions may increasingly deploy decoys that seek out fraud, gather evidence, and disrupt campaigns before losses mount.

In that context, tools like Apate are important not because they solve every case, but because they show what a more active defense model could look like. Rather than treating scams as an endless flood to be endured, the field is beginning to ask whether the flood can be slowed by making every drop more expensive for the attacker.

That is the strategic shift at the heart of this story. Artificial intelligence helped scammers scale faster. Now it is helping defenders scale back.

Key facts at a glance

The practical lesson for consumers is unchanged: suspicious calls and texts remain dangerous, even as defenders get more sophisticated. But behind the scenes, the anti-fraud industry is changing fast. AI is becoming both the problem and, increasingly, part of the solution.

That may be the most important development in the fight against cybercrime right now. The same technology that helps criminals scale can also be used to drain their time, expose their methods, and limit their reach. If the new scam economy runs on speed, AI is emerging as one of the few tools that can force it to slow down.

How are scammers being fought with AI right now?

Scammers are being fought with AI by using bots and honeypots that impersonate real victims, respond to messages, and keep fraudsters engaged while collecting intelligence. Companies like Apate and some researchers also use large language models to make traps more realistic and harder to detect.

Who is Apate and what does it do?

Apate is an Australian company that builds AI bots to answer scam calls, reply to texts, and infiltrate scam chat groups. It says the system is used by banks and supported by telecoms to delay criminals and collect information about fraud operations in real time.

Why does keeping scammers busy matter?

Keeping scammers busy matters because fraud operations rely on speed, volume, and repeated contact attempts. If an AI bot can hold a scammer in conversation, it can reduce the number of real people reached, while also creating a record of the scammer’s methods and infrastructure.