Apate.ai Uses AI Decoys to Tie Up Scammers and Collect Fraud Data
Apate.ai deploys thousands of autonomous AI agents to pose as scam victims, keep fraud operators engaged, and extract operational intelligence for institutional customers.
The company is an Australian startup founded in 2024 from research at Macquarie University in Sydney. It is led by founder and CEO Dali Kaafar with co-founder and chief commercial officer Brad Joffe. Kaafar traces the idea to keeping a scammer on the phone for 44 minutes at a family picnic TechCrunch.
Instead of only filtering malicious calls and messages, Apate sends out decoys that answer them. Those decoys are software programs that can hold a conversation on their own, without a human in the loop.
Those decoys answer scam calls, respond to messages and emails, follow phishing links, which are fake links designed to steal logins, and join social media groups to bait fraudsters TechCrunch. The agents are described elsewhere as lifelike agents that engage scammers AFP. The company says scale is what sets it apart. A human scam-baiter can tie up one operator at a time. A fleet can hold thousands of parallel sessions, meaning many conversations at once.
Turning talk time into telemetry
What the agents collect during those sessions is the product. The fleet aims to capture scammers' bank accounts, crypto wallets, which are digital accounts for cryptocurrency, phone numbers, malicious websites, impersonated brands, and scam details TechCrunch.
The company says the fleet has wasted 17 years of scammers' time across 2.5 million conversations TechCrunch. That figure is self-reported. It works both as a measure of time wasted and as a sign of how large its store of conversations has grown. Every minute held is also another turn of dialogue, another payment instruction, another infrastructure artifact, such as a technical trace, to log.
On its site, Apate states it detects scam calls and texts at the source and engages attackers before they reach real victims Apate. It states its bots disrupt scam conversations in real time by misleading scammers and derailing fraud attempts Apate. It also states it captures tactics and attack vectors, the methods attackers use, as they unfold, including email addresses from scam attempts as actionable intelligence, and provides customisable dashboards, alerts and analytics built for decision-making Apate.
In vendor terms, the company is described as an autonomous intelligence and disruption platform built to fight fraud at scale AFP.
A distribution play through telecom and finance
Apate's customers include banks, telecom companies, government agencies, and utilities, which use its intelligence for anti-fraud measures TechCrunch. Those organizations carry the traffic, hold the accounts, or face the losses and regulatory pressure.
For communication service providers, the company partnered with Mobileum to turn scam interactions into actionable intelligence Apate. According to the companies, Mobileum and Apate's AI-powered conversational agents help providers go beyond call and message blocking to disrupt fraudsters before they reach real customers Press Release Hub.
On the business side, Apate.ai raised US$8.15 million to take its autonomous fleet of AI agents to a global fight against scammers and reincorporated in the United States Apate. It pledged to support fraud loss prevention through intelligence-led scam disruption deployments by March 2027 UNODC. That pledge includes a commitment that by March 2027 it will deploy AI-driven conversational agents to proactively engage fraud actors across voice channels UNODC.
Apate was a finalist in the Startup Battlefield 200 at TechCrunch Disrupt 2026 TechCrunch.
Where the leverage actually sits
The broader context here is a shift from blocking at the edge to collecting at the source. Blocking a number or web address is cheap for an attacker to work around. Getting the attacker to volunteer a fresh mule account, an account used to move stolen money, a wallet, or a phishing-kit web address produces traces that can feed detection models, account closures, and takedowns across multiple defenders at once.
In my view, the technical bet is that conversational persistence is now good enough to be operationally useful. A decoy does not need to fool a language expert. It needs to stay coherent through credential requests, payment pivots, and urgency scripts, without breaking character or revealing that it is automated. Anyone who has watched teenagers test the limits of a chatbot will recognize the dynamic. Scam operators will probe, improvise, and adapt just as quickly.
Looking at what this means for buyers, the value comes down to feed quality and workflow fit. Raw transcripts do not stop losses. Timely, de-duplicated indicators, cleaned-up warning signs with no repeats, with context on impersonated brand, vector, the channel used, and operator behavior, delivered into fraud risk engines and network abuse teams, might.
Worth flagging here is the operational risk that comes with engagement. Following phishing links, joining groups, and holding voice sessions at scale requires strict containment, careful handling of personal data that surfaces in transcripts, and clear boundaries with law enforcement partners.
The optimistic case here is that this changes the economics in a useful way. Fraud operations depend on high-volume outreach and low-friction conversion. If a meaningful share of responses are machines that waste operator time while mapping infrastructure, the cost per successful victim rises. That will not end scamming. It could make specific campaigns and kits more expensive to run, and give banks, carriers, and agencies earlier warning of what is circulating. For defenders who have spent years working with blocklists against rotating numbers and domains, that earlier signal is the part to watch.
