Executive brief
A security flaw in Apple M1 processors allows a malicious app to steal sensitive data from other apps running on the same computer. By exploiting how the graphics processor (GPU) handles temporary memory, an attacker can read "leftover" information—such as passwords or encryption keys—that was recently used by a different program. This bypasses standard security protections that are supposed to keep apps isolated from one another.
Technical details
The vulnerability is a cross-process information disclosure (CWE-459) occurring in the Apple M1 GPU. The hardware fails to zero out or isolate register file state between compute shader dispatches originating from different execution contexts. A local attacker can deploy a sandboxed Metal application that executes a 'reader' shader to capture stale register values left behind by a victim process. Proof-of-concept code demonstrates the successful recovery of a 128-bit secret generated by a separate sandboxed process without any IPC or shared memory. While Apple considers this 'legacy' behavior addressed in newer silicon, the flaw remains present in M1-based hardware.
Affected products
- Apple M1 Silicon Systems using M1 chips (e.g., MacBook Air 10,1) running macOS 26.3.1 and earlier
Timeline
- 2026-05-29: other: Last active date of the technical disclosure gist
- 2026-06-24: disclosed: CVE published as a disputed record