Apple Intelligence researchers detailed patched prompt injection bypass

TL;DR:

• Researchers detailed a prompt injection attack that bypassed Apple Intelligence protections on April 9, 2026.
• The issue let attackers circumvent restrictions and force the on-device large language model to execute attacker-controlled actions.
• Apple has corrected the issue, according to the report.

Researchers detailed a prompt injection attack that bypassed Apple Intelligence protections on April 9, 2026. The report said the now corrected issue let attackers circumvent Apple’s restrictions and force the on-device large language model to execute attacker-controlled actions.

The report described the issue as a prompt injection attack against Apple Intelligence’s on-device large language model. According to the source, the attack bypassed Apple’s protections and caused the model to carry out attacker-controlled actions before Apple corrected the issue.

The report adds to wider scrutiny of prompt injection risks in AI systems, including on-device models that are designed to operate with built-in safeguards. In this case, the source said researchers were able to bypass Apple Intelligence protections, and Apple has since corrected the issue.

Related Questions

Was Apple Intelligence vulnerable to a prompt injection attack?

Yes. Researchers said a now corrected issue allowed a prompt injection attack to bypass Apple Intelligence protections and force the on-device model to execute attacker-controlled actions.

Did Apple fix the Apple Intelligence prompt injection issue?

Yes. The report said the issue had been corrected after researchers detailed how it bypassed Apple Intelligence protections.