First Apple M5 Kernel Vulnerability Revealed: Five‑Year MIE Defense Cracked in 5 Days

Event Overview

The Calif research team visited Apple’s Cupertino headquarters and submitted a report showing the world’s first macOS kernel memory‑corruption vulnerability running on the M5 chip, successfully bypassing Apple’s flagship Memory Integrity Enforcement (MIE) protection.

Technical Background

MIE is Apple’s hardware‑level memory‑safety system built on ARM’s Memory Tagging Extension (MTE). Apple invested roughly $5 billion and five years in its development and claimed it could resist all known public exploit chains, including the Coruna and Darksword toolkits.

Vulnerability Discovery Process

Bruce Dang identified the initial flaw on April 25. Dion Blazakis joined the effort on April 27, and Josh Maine constructed the exploitation toolchain. By May 1, the complete exploit chain was operational, meaning the entire process took only 5 days .

Exploit Chain Details

Target platform: macOS 26.4.1 (25E253) running on bare‑metal M5 hardware

Vulnerability type: Data‑Only Kernel Local Privilege Escalation (LPE)

Attack start point: ordinary non‑privileged local user

Attack method: uses only normal system calls

Final effect: obtains a root shell

The chain combines two independent vulnerabilities and several auxiliary techniques to achieve code execution on an M5 device with MIE enabled.

AI’s Key Contribution

During development, the team leveraged Anthropic’s security research platform Mythos Preview . Mythos helped identify the vulnerability and participated throughout the exploit development, demonstrating the ability to generalize an attack technique across similar problems. However, because MIE is a brand‑new top‑tier mitigation, human security experts were still essential.

Research Significance

The core goal was to explore what can be achieved when state‑of‑the‑art AI models collaborate with human security experts. Completing a kernel memory‑corruption exploit against a top‑tier defense in a week highlights two important points: (1) no mitigation is unbreakable, and (2) AI is rapidly lowering the barrier for developing high‑complexity exploits.

"MIE is not invulnerable. With the right bug, attackers can fully bypass it. As AI systems discover more vulnerabilities, some will become powerful enough to break even advanced mitigations like MIE."

The team plans to publish a 55‑page technical report after Apple releases a patch.

Security Implications

Memory safety remains critical

Memory‑corruption bugs are the most common vulnerability class on iOS and macOS. Even massive hardware‑level investments cannot guarantee absolute security.

AI is reshaping the threat landscape

Calif’s success shows that AI dramatically reduces the development effort required for sophisticated exploits, enabling small teams to accomplish work that previously required large organizations.

"Apple built MIE in a pre‑AI world. We are about to witness how the best mitigations fare in an era of AI‑driven vulnerability explosions."