Apple Intelligence Prompt-Injection Flaw Exposed With 76% Success Rate

Apple Intelligence is Apple's core service for integrating generative AI into the iPhone, iPad and Mac. Prompt injection uses specially crafted instructions to mislead a model into ignoring its original rules. Researchers said such attacks could bypass content guardrails and induce inappropriate responses, highlighting data and user-safety risks as AI moves onto consumer devices.

A new study disclosed a prompt-injection technique targeting Apple Intelligence that achieved an attack success rate of up to 76% in testing, bypassing Apple's safeguards and manipulating model output. Apple has patched the flaw in newer operating-system releases, including iOS 26.4. The incident involved no financial losses, but experts warned that model updates could create new avenues for bypasses, requiring AI defenses to be tested continuously.