Source: The Verge
Anthropic and other researchers have documented active exploitation of large language models in dating fraud at scale—thousands of victims targeted with AI-generated personae that mimic authentic romantic interest. Bad actors are monetizing LLM conversational fluency and narrative coherence to conduct social engineering and financial theft. The finding exposes a gap between Anthropic's safety positioning and real-world vulnerability: the company can publish disclaimers about misuse, but has limited enforcement mechanisms once models are deployed or accessed through third-party integrations.