Apple Intelligence Rewrites Its Own Rules After One Sneaky Prompt
Ourtake
Apple's guardrails held up better than most, but the 'ignore previous instructions' trick is still undefeated since GPT-3.
What happened
A developer showed that prompt injection could override the system prompt of Apple Intelligence's Rewrite feature in MacOS 15.1 Beta 1, making it respond to arbitrary prompts. The developer published the code on GitHub after a YouTube demonstration.