On-device Apple Intelligence vulnerable to prompt injection techniques
AI-summarised brief · reviewed before publication
Researchers from RSAC Research have discovered a vulnerability in Apple's on-device AI, allowing attackers to manipulate it using prompt injection techniques. The method, which achieved a 76% success rate in 100 tests, employs adversarial prompts and Unicode obfuscation. This could grant access to sensitive user data. The findings were shared with Apple on October 15, 2025, highlighting a potential security risk in Apple's operating systems.