Apate uses undercover AI agents to collect scam intelligence

Apate.ai is using AI agents for a job that looks very different from sales outreach or back-office automation: pretending to be scam victims. According to TechCrunch, the Australian startup sends thousands of undercover agents into calls, messages, emails and online spaces to keep fraudsters talking and extract useful intelligence.

For banks, telecoms, utilities and government agencies, the pitch is simple: learn how scams work before a real customer or employee is harmed. The approach also shows how autonomous agents are moving into adversarial workflows, where the goal is not just to complete a task, but to safely interact with bad actors.

Undercover agents, not productivity bots

Apate was founded in 2024 and is based on research carried out at Macquarie University in Sydney, TechCrunch reports. The company’s AI agents are designed to look like plausible targets. They can answer scam calls, reply to suspicious texts and emails, follow phishing links and enter social media groups.

The agents vary their apparent identities, including age, gender, accent and language, according to the report. Their purpose is to hold scammers’ attention long enough to reveal operational details that defenders can use.

Apate says its agents have already consumed 17 years of scammers’ time across 2.5 million conversations. The more important output, the company argues, is not the distraction itself but the data gathered during those conversations.

That data can include:

• bank account numbers and crypto wallet addresses
• phone numbers used by scammers
• malicious websites and phishing links
• brands being impersonated
• details about the type of fraud being attempted

The company then provides that intelligence to customers, who can apply it to anti-fraud efforts. Apate is also a finalist in TechCrunch Disrupt’s Startup Battlefield 200, with the event taking place this week at Moscone West in San Francisco through Thursday, October 15.

Why pre-victim fraud data matters

Most anti-fraud systems respond after a signal appears: a suspicious transaction, a reported phone number, a known phishing site or a customer complaint. Apate’s bet is that AI agents can move some of that work upstream by engaging scammers directly and gathering indicators before a victim reaches the final step.

Brad Joffe, Apate’s co-founder and chief commercial officer, told TechCrunch that telcos may block numbers after the fact and banks may detect transactions after the scammer has already reached the customer. Apate’s model is to interact with the scammer earlier and return the resulting information to customers.

The timing is notable because the same class of AI tools is also being used by criminals. According to FBI figures cited in the TechCrunch report, the bureau received more than 22,000 complaints about AI-related crimes in 2025, tied to about $893 million in losses. Investment scams accounted for more than $632 million of that total, with the FBI saying fraudsters use AI to conduct conversations and deepfake trusted figures.

For small businesses and creators, the lesson is not that everyone should build scam-baiting bots. It is that fraud defense is becoming more active and agent-based. Teams designing their own anti-scam workflows need to think about what an agent is allowed to do, what data it can collect, and how human reviewers verify findings before acting on them.

How Apate’s system is used by customers

Joffe told TechCrunch that Apate uses both frontier models and its own fine-tuned models for the agents and analysis. The company trains models on conversations its agents have had with scammers, as well as earlier research from founder and CEO Dali Kaafar’s work at Macquarie University.

Customers receive access to Apate’s agent fleet and an insights portal with analytics on scams targeting their users, employees, brands and products, according to the report. Customers can also give agents specific missions, such as trying to obtain a bank account number or crypto wallet address.

The text-based agents can look through forums, investment groups and social platforms for possible scam activity, then join conversations using a fabricated identity to surface details about the operation.

TechCrunch reports that Apate’s customers include Commonwealth Bank of Australia and TPG Telecom, alongside other utilities, telecom companies and government agencies. Joffe said TPG has routed nearly 590,000 calls to Apate’s agents. He also said Commonwealth has seen a 76% drop in publicly reported scam losses, and that an unnamed Australian bank working with Apate used outbound agents to disrupt 22 scam campaigns.

Those are customer-specific claims reported by TechCrunch, not a general benchmark for the market. Still, they point to a practical pattern: AI agents can be used as collection points for fraud intelligence, while existing security, compliance and customer-protection teams decide how to respond.

For readers building automation around trust and safety, this sits near the same broader trend as AI cyber-defense tools such as Artemis Security’s Orion-1: software is taking on more investigative work, but the consequences of errors remain human and financial.

What to watch next

Apate raised $8.15 million in seed funding in August from Lobby Capital, OIF Ventures, Investible, Concept Ventures and Baobab Ventures, according to TechCrunch. Joffe said the 20-person company has raised just under $10 million in total.

The startup plans to use the latest round for product work and international expansion, including opening an office in London and expanding its U.S. presence, the report says.

The larger question is how organizations will govern agents that intentionally enter hostile environments. Anti-scam agents may need clear mission limits, evidence handling, audit trails and escalation paths. For teams already mapping automations across support, security and compliance, our automation buying guide is a useful reminder that workflow design matters as much as the model behind it.

Apate’s story suggests a new category of defensive automation: agents that do not merely wait for threats to arrive, but go looking for them under controlled conditions.