随著网路诈骗日益猖獗且跨境打击难度极高，澳洲新创公司 Apate 开发了一套名为「完美受害者」的 AI 机器人系统，专门将诈骗通话导流至虚拟受害者手中。该平台部署了约 35 万个具有不同性格与语言能力的机器人，旨在长时间牵制诈骗分子以消耗其时间与资源，同时即时搜集诈骗集团的 URL、人头帐户及银行明细等珍贵情报。

Apate 的 AI 人设表现出适度的怀疑与配合，给予骗子希望却绝不上当，实际测试显示对话节奏极为自然且具说服力，通话往往能持续数小时之久。尽管诈骗集团每年发动数十亿次攻击、难以一夕根除，但生成式 AI 在反诈骗领域展现出强大潜力，能借由心理战术直接牵制犯罪者，并大幅降低潜在真实受害者接到诈骗电话的风险。

除了语音诱捕，安全领域也正广泛将大型语言模型整合至蜜罐系统（Honeypots）中，苏黎世联邦理工学院的研究即指出，LLM 驱动的蜜罐能展现更逼真的行为，使攻击者耗费显著更长的时间且更难察觉受骗。透过生成式 AI 消耗骇客资源并搜集情报，正在为数位安全防御者开创对抗网路犯罪的全新优势。

As cybercrime escalates and cross-border enforcement remains challenging, Australian startup Apate has developed an AI system designed to act as 'perfect victims' by diverting scam calls to automated bots. Operating around 350,000 bots with diverse personas and language capabilities, the platform engages fraudsters in protracted conversations to waste their time and resources while harvesting real-time intelligence on scam URLs and mule accounts. (Key numbers: 35)

Apate's AI personas exhibit a calibrated mix of skepticism and gullibility, keeping scammers hopeful without ever falling for the trap, with tests demonstrating natural dialogue flow that can prolong calls for hours. Although cybercriminals continue to initiate billions of attacks annually, generative AI shows immense promise in frustrating scam operations and significantly curbing the number of potential human victims reached by automated dialers.

Beyond voice baiting, security researchers are increasingly integrating large language models into honeypot systems, with studies from ETH Zurich showing that LLM-simulated honeypots interact far more realistically and detain attackers for significantly longer periods. By actively draining adversaries' resources and gathering actionable data, generative AI offers defenders a potent new edge in disrupting cybercrime.