Defenders Deploy Thousands of Artificial Intelligence Decoys to Drain Cybercriminals of Time and Money Security innovators are turning conversational bots and language models into realistic targets that tie up phone scammers for hours while gathering vital operational intelligence. As digital fraudsters harness modern automation to launch industrial-scale operations across the globe, cybersecurity researchers are countering with a mirrored strategy: turning artificial intelligence into a tireless decoy. Law enforcement agencies have repeatedly encountered severe jurisdictional limits when tracking digital crime syndicates that operate safely beyond sovereign borders. Rather than relying solely on traditional cross-border crackdowns, engineers are building autonomous defensive systems designed to absorb incoming scams, waste criminal labor, and disrupt illicit business models from within. Automated Victims Designed to Waste Scammer Labor Over the past two years, Australian firm Apate has been constructing a specialized infrastructure that diverts fraudulent phone traffic directly to artificial intelligence agents. The company, named after the ancient Greek personification of deceit, trains its conversational systems to prolong interactions for as long as possible without ever compromising real credentials or funds. The core objective is to feed fraudsters just enough optimism to keep them engaged on the line, using that extended engagement window to harvest critical data about ongoing schemes. Dali Kaafar, the founder and chief executive officer of Apate, summarized the strategic purpose behind this architecture: "What we really like to think is that we’re building the perfect victims for scammers." He explained that every minute a criminal spends negotiating with an automated agent is a minute where potential targets are spared from robo-dialing sweeps that can contact thousands of households simultaneously. Hundreds of Thousands of Personas Intercepting Data Supported by telecommunications carriers and integrated into banking risk frameworks, Apate operates an active deployment of roughly 350,000 conversational bots. These synthetic personas do not simply pick up suspicious inbound voice calls; they also enter clandestine messaging channels and answer incoming fraudulent SMS traffic. Through this persistent interception, the platform has compiled more than 250,000 distinct intelligence points, identifying fraudulent payment gateways, active money mule networks, and operational bank accounts in real time. To maintain credibility across varied cultural contexts, the bots are programmed with diverse accents, communication habits, and digital footprints. Dali Kaafar pointed out that the personas behave with the natural inconsistency of genuine human beings. Some maintain active WhatsApp profiles while others do not, and certain agents will answer immediately whereas others abruptly hang up after promising to call back later. This measured irregularity prevents callers from realizing they are interacting with algorithmic counterparts. Prolonged Engagements and Behavioral Friction Evaluations of the system show that the agents project an authentic layer of healthy skepticism while leaving sufficient behavioral openings for callers to persist. This subtle balance creates immense conversational friction for the caller, yet it mirrors the exact hesitations scammers routinely encounter from cautious consumers. Consequently, interactions handled by Apate routinely exceed two continuous hours on single voice calls. During an experimental test involving a simulated cryptocurrency opportunity, two human callers acting in tandem as a client and an investment advisor failed to extract a financial commitment from the system after six minutes of sustained pitching. Despite promises of extraordinary financial returns, the persona engaged smoothly with realistic timing and natural pauses while maintaining firm financial boundaries. The demonstration underscored how effectively fine-tuned conversational models can mimic human dialogue without succumbing to social engineering. Expanding From Decoy Calls to Advanced Honeypots Despite these technological strides, the sheer volume of illicit communication remains overwhelming, with international syndicates transmitting billions of fraudulent messages and maintaining sprawling physical facilities. Longstanding efforts by investigative authorities and independent baiters have offered vital friction, but systemic eradication remains elusive without deep, continuous data sharing across banking, telecommunications, and law enforcement sectors. The concept of diverting malicious actors into resource-wasting traps is expanding across other domains of digital security, notably through honeypots. Organizations have long deployed simulated virtual servers to lure unauthorized intrusions, study novel penetration tools, and exhaust intruder resources. Mark Vero, a doctoral researcher in the computer science department at ETH Zurich, indicated that open source honeypot frameworks are integrating large language models to deliver vastly superior behavioral authenticity. Recent investigations conducted by Vero and his team revealed that language-model-driven honeypots hold automated offensive agents for substantially longer durations than legacy traps with static responses. Mark Vero observed that automated penetration systems demonstrate far lower detection rates against adaptive models, confirming that deploying responsive artificial intelligence environments offers a tangible strategic edge to defensive teams seeking to frustrate bad actors. What this means for you Integrating conversational artificial intelligence into telecom and banking backends directly reduces the operational bandwidth available to criminal call centers. • Call Volumes: Telecom providers are diverting automated fraudulent calls into synthetic listening loops. This ties up outbound lines and prevents criminals from reaching thousands of potential consumer targets. • Financial Safeguards: Real-time intelligence collection reveals active money mule networks and compromised banking routes. Financial institutions can freeze illicit transaction channels before funds disappear overseas. • Consumer Vigilance: Despite defensive bots absorbing heavy traffic, individuals must remain disciplined regarding personal data. Never share one-time passwords, private credentials, or remote device access with unknown callers. • Threat Mitigation: Captured fraudulent scripts and malicious website domains feed directly into threat intelligence registries. This allows security filters to block scam links on mobile platforms much faster. Why this happened International jurisdictional friction and the widespread adoption of mass-dialing software by syndicates compelled defenders to develop automated mechanisms that erode the economic viability of scamming. • Jurisdictional Hurdles: Large criminal networks operate primarily from overseas enclaves beyond the reach of local law enforcement. Pursuing physical arrests across borders involves complex diplomatic and procedural delays. • Industrialized Dialing: Syndicates deployed automated robocall tools capable of contacting millions of phone numbers concurrently. Human response teams could not match that volume without autonomous conversational agents. • Economic Disruption: Forcing fraud networks to spend expensive labor hours on unproductive artificial personas degrades their profitability. Making attacks financially unviable reduces the overall scale of outbound criminal campaigns. Questions & Answers 1. How does Apate's defensive system operate? The platform diverts suspected scam calls to conversational bots that emulate cautious human victims, keeping fraudsters engaged while collecting operational intelligence. 2. How many artificial personas does the platform deploy? The infrastructure utilizes approximately 350,000 distinct bots across phone and digital messaging channels. 3. What intelligence has been harvested through these interactions? The system has recorded more than 250,000 pieces of actionable data, including fraudulent links, mule accounts, and banking coordinates. 4. What did research at ETH Zurich show regarding honeypots? The research demonstrated that honeypots powered by large language models detain automated attackers significantly longer than traditional static traps. https://trendkia.com/en/security/saibara-thagon-ko-ulajhane-ke-lie-suraksha-ejensiyan-aba-ai-botsa-ka-le-rahin-sahara-45997 TrendKia — Har trend, sabse pehle.