200,000 Fake AI Victims Used to Lure and Expose Online Scammers

AI scam-baiting has moved from novelty to infrastructure. Australian company Apate says it has deployed nearly 200,000 AI “persona” characters worldwide that impersonate gullible targets—keeping fraudsters busy while simultaneously collecting intelligence for banks and telecom operators.

In a recent six-week window ending in 2025, Apate reported that its bots engaged in 600,000 scam calls for Australian telco TPG, which the company frames as more than 500 days of wasted criminal time—equivalent to savings of roughly $13 million. Beyond disruption, Apate’s system is designed to surface actionable details that can help defenders identify where stolen funds are moving and how scam networks operate.

Key takeaways

• Apate says it runs close to 200,000 AI personas that can interact with scammers via phone-style conversations and messaging channels.
• In the six weeks to late 2025, Apate reported 600,000 scam calls engaged for TPG, translating into hundreds of days of scammers’ time lost.
• The company’s anti-scam value proposition is not only delay—it also centers on extracting intelligence such as new cryptocurrency wallet addresses.
• Apate’s research suggests a growing share of scam communications already involve AI, but it argues defenders can still have an advantage.
• The approach mirrors broader industry efforts, including UK telco O2’s “AI Granny” campaign, but Apate positions its scale and data extraction as the differentiator.

AI targets built to waste scammer time

Apate founder Dali Kaafar describes the system as a way to automate what human scam baiters have long attempted manually: stringing fraudsters along to consume their effort, not the public’s. In Kaafar’s telling, the idea emerged after he personally received a scam call while on a family picnic in Sydney in November 2021. He spent 44 minutes engaging the caller by roleplaying as a naive victim.

What began as a private diversion evolved into a research-led project. While working as a professor at Macquarie University, Kaafar discussed the concept with doctoral students focused on AI and security, proposing a system that could both engage scammers at scale and capture useful information from those interactions.

In a matter of months, the project secured funding from the Office of National Intelligence for research work before spinning out into Apate in 2023. Kaafar says the company now works with major banks in Australia and with other financial institutions in the UK, South Africa, and parts of Southeast Asia.

Nearly 200,000 personas and “realistic” conversation behavior

Apate launched with 120 distinct personas and later expanded to 197,000. Kaafar attributes realism to detailed characterization, including identifiable vocal tics, accents, and small behavioral cues. He says the company spent considerable time refining how the bots sound and respond so that conversations feel natural to targets—and convincing enough for scammers who may be skeptical.

According to Kaafar, the underlying AI models were trained on “hundreds and hundreds” of recorded conversations between human scam baiters and scammers. This training is aimed at enabling counter-strategies during calls and chats, rather than simply running automated scripts.

Apate also uses the same engagement loop as a measurement tool. Kaafar says one internal performance metric tracks the frequency of profanity directed at the bots by frustrated scammers—an anecdote that underscores how the company is optimizing for sustained engagement rather than quick hang-ups.

The company deploys the bots through channels including WhatsApp and Telegram, where scammers often try to move fast from initial contact toward payment instructions. Kaafar also frames the strategy around a key behavioral truth: even if “you can’t scam an honest man” is not literally correct, fraudsters remain motivated by greed, and that motivation can still be exploited by delaying or steering their workflows.

From disruption to defense: extracting crypto and operational intelligence

Apate describes its anti-scam data goal as forward-looking intelligence for banks and telecom operators. In its crypto-focused work, Kaafar says Apate partners with “one of the leaders in blockchain analysis” and is interested in identifying wallet addresses and methods used by scam rings.

Kaafar claims that, during bot engagements across multiple conversations, Apate can extract new crypto wallet addresses “by the hundreds and by the thousands.” The implied rationale is that defenders need to identify the next place where money will land before funds are transferred—so monitoring and incident response can be applied ahead of damage.

He compares scam operations to corporate organizations, suggesting that call centers and related workflows are structured enough to support a repeatable data advantage. In this framing, the most important output from a bot interaction is not merely evidence that fraud occurred, but the details that help analysts understand which accounts, wallets, and compounding “money collection” points matter next.

As an example, Apate says that in July its bots uncovered a marketplace involving brokers soliciting verified bank accounts in India, with commissions reportedly paid in USDT based on proceeds from scams passing through those accounts.

An arms race where defenders may still have an edge

Apate’s broader warning is that scammers are increasingly using AI too. The company cites that scams can scale cheaply because fraud is already a large business. Kaafar says Apate’s research estimates that about 20% to 30% of scam text conversations already employ AI, reflecting how quickly fraudsters can adopt tools to accelerate communications.

Still, Kaafar argues that anti-scam bot systems have a strategic advantage in a defender-vs-attacker AI setting. Drawing on game theory concepts, he suggests that defensive bots are built to extract information, while scam bots are trying to push the other side toward an action—meaning the defender can more easily learn from the attacker’s model and behavior.

Kaafar also takes a longer view: even if scammers become more sophisticated, the same sophistication may increase the amount of exploitable data left behind in the interaction. He describes that dynamic as “good news” in the fight against scams.

The point is not that AI eliminates fraud risk, but that well-designed engagement systems can convert fraud attempts into intelligence streams—turning what would otherwise be wasted time for victims into a resource for investigators and monitoring teams.

As AI scam methods evolve, readers should watch whether bot-based intelligence extraction becomes standard among financial institutions and telecom providers—and, crucially, how quickly defenders can operationalize the wallet and account details that these systems surface ahead of transfers.