Researchers Find LLM API Keys Exposed in 282 iOS AI Apps

Researchers at Wake Forest University dynamically analyzed 444 iOS apps with LLM features and found that 282 of them (64%) exposed LLM API credentials or backend access over network traffic. Of those, 146 (52%) were confirmed to be fully exploitable, raising the risk of "LLMjacking," in which a third party runs costly LLM inference for free using someone else's keys. Details of the research