Cybercrime has turned into a full-blown AI arms race, and for once, the good guys might be getting the more interesting toys. As first reported by Wired, anti-scam outfits are now deploying lifelike AI bots that pretend to be victims, string fraudsters along for minutes or hours, and quietly harvest data that can be passed to investigators and platforms to dismantle scam networks.
One of the most striking examples comes from Apate, an Australian firm named after the Greek goddess of deceit that has spent the past two years building an AI system to intercept phone scams and route them to synthetic “victims.” When a scammer dials a targeted number, Apate’s infrastructure can redirect the call to a bot that sounds and behaves like a real human, complete with different personalities, languages, and backstories tuned to keep the scammer engaged without ever actually falling for the con. The goal is twofold: waste criminals’ time so they can’t reach real targets, and extract useful intelligence—phone numbers, money-mule accounts, scripts, and social engineering techniques—that can feed law enforcement and corporate security teams.
This “fight fire with fire” approach is spreading beyond one startup. BeeSafe AI, backed by Y Combinator, is pitching a platform that explicitly engages scammers instead of just blocking them, running thousands of real-time conversations across email, SMS, messaging apps, and social media to map out the infrastructure behind trust-based scams. According to the company, its bots are already helping banks, telecoms, and government agencies identify mule accounts, fraudulent domains, crypto drainers, and other assets that keep modern fraud operations running, effectively turning every intercepted scam message into a potential tip-off about the larger crime ring behind it.
Big platforms are quietly weaponizing AI against fraudsters too. Meta has detailed how it uses advanced machine-learning systems to analyze text, images, and contextual signals to catch impersonation scams, deceptive links, and domain spoofing across its apps. Those tools can now flag suspicious job offers or stranger DMs, ask users if they want an AI “scam review,” and, when the patterns match known attacks, suggest blocking or reporting the account. In parallel, tools like Charley from Charlemagne Labs monitor incoming messages to warn users about romance scams, phishing, and other social engineering attempts, while social-engineering specialists such as Rachel Tobac of SocialProof have been sounding the alarm that scammers are already using AI to clone voices, generate convincing emails, and spin up deepfake videos of public figures.
That last point is the twist: defenders may be trolling scammers with bots, but criminals are building their own dark-mirror versions of consumer AI. Reporting from Wired has documented underground models like “FraudGPT” and “WormGPT,” language-model spin-offs that are marketed in cybercrime forums as tools for crafting polished phishing campaigns, writing malware, and automating social engineering. Another investigation showed how multiple mainstream and open-source AI models can be prompted to design disturbingly effective phishing messages and attack chains, underscoring how easily generative systems can be misused when guardrails are weak or bypassed. NPR’s reporting on global scam compounds—where trafficked workers are forced to run digital fraud operations—has further highlighted how generative AI lets those groups instantly churn out flawless, personalized bait emails at massive scale.
Scam content itself is getting a visual upgrade. Separate Wired coverage has tracked the rise of “AI face models,” with real-world influencers applying to have their likeness turned into endlessly reusable deepfake avatars for marketing—avatars that scammers can co-opt for fake investment pitches, romance scams, or phony celebrity endorsements. Combined with voice cloning and synthetic video, this means anti-scam AI is increasingly chatting with bots on the other side too: lifelike scam agents fronting for human crews, and human-sounding defensive bots dragging those crews into dead-end conversations. It’s a weird, thoroughly cyberpunk moment where vast botnets of NPC-like agents are arguing about bank transfers and password resets, all while human operators watch dashboards and tweak scripts.
There are real risks in turning AI loose on scammers, from false positives that rope innocents into bot conversations to broader concerns about surveillance and data sharing across banks, platforms, and governments. Even so, the consensus in recent reporting is that defenders need automated help to keep up with AI-enhanced fraud, and that sophisticated information sharing—powered by AI analysis of scam patterns, infrastructure, and social graphs—could make a real dent in global cybercrime if it’s deployed with transparency and legal oversight. For now, the message to cybercriminals is clear: every email, DM, and phone call might already be talking to a machine, and that machine isn’t just stalling—it’s quietly looting their playbook for the next level of the fight.
