University of California researchers have discovered that some third-party AI large language model (LLM) routers can pose security vulnerabilities that can lead to crypto theft.  A paper measuring malicious intermediary attacks on the LLM supply [...]