Apple Intelligence AI Guardrails Bypassed in New Attack

Researchers from RSAC discovered a way to bypass Apple Intelligence's safety protocols with a 76% success rate by combining two attack techniques: Neural Execs, which uses gibberish inputs to trick the AI, and Unicode manipulation, which uses backward text with right-to-left override functions to evade content filters. The attack could potentially allow malicious actors to manipulate private data in third-party apps integrated with Apple Intelligence, affecting an estimated 100,000 to 1 million users, though Apple has since rolled out protections in recent iOS and macOS updates. RSAC estimates there were 200 million Apple Intelligence-capable devices in users' hands as of December 2025, making it a high-value target for attackers.