1 article
A research team bypassed Apple Intelligence's on-device LLM guardrails using prompt injection.