Anthropic's AI model, Claude Mitose Preview, has successfully compromised Apple's latest security technology, which the company invested billions in over the past five years, in just five days.
According to the IT industry on May 17, the Vietnamese security research firm Caliph announced on May 14 via a Substack blog that it had developed the first-ever macOS kernel memory corruption exploit targeting Macs equipped with the Apple M5 chip.
The security technology breached by Caliph is Apple's Memory Integrity Enforcement (MIE), a hardware-based memory safety system implemented exclusively by Apple on its M5 and A19 chips. MIE operates by assigning a 4-bit tag to each 16-byte memory block and immediately blocking access if there is a tag mismatch.
The research team utilized Mitose to first discover the bug on April 25 and secured the completed attack code by May 1. The exploit begins with standard user privileges and escalates to the highest system privilege, known as 'root,' by chaining two vulnerabilities to target the macOS kernel.
The team visited Apple's headquarters to report the vulnerability directly. Caliph stated that they chose in-person reporting to avoid being lost in the flood of bug submissions. Following this, Apple acknowledged Caliph and Anthropic Research as contributors to the vulnerability fix in the release notes for the macOS Tahoe 26.5 update.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.