Security researchers tricked Apple Intelligence into cursing at users. It could have been a lot worse

2026-04-10

RSAC researchers hijacked Apple Intelligence on supported iPhones using a Neural Exec prompt‑injection attack plus a Unicode right‑to‑left override trick. They encoded malicious English output backwards, bypassed Apple's filters, and got 76 of 100 prompts to execute, including abusive replies and silent actions like creating or renaming contacts to gain trust on a victim's device.