Researchers find way to bypass Apple’s on-device LLM safeguards

Researchers have found a way to bypass Apple's restrictions on its on-device LLM, successfully executing attacker-controlled actions through prompt injection.