
Researchers from UC Santa Barbara, UC San Diego, Fuzzland, and World Liberty Financial discovered that LLM routers—intermediary services between users and AI models—pose significant security risks. These routers can steal credentials and drain cryptocurrency wallets. One malicious router drained a client's $500,000 wallet. Researchers identified 26 routers injecting malicious commands. As AI agents are projected to mediate trillions in commerce by 2030, this vulnerability threatens widespread crypto transactions.