Chip Apple The M4 hides a secret. The developer bypassed its limits – LsA magazine

Chip Apple The M4 hides a secret. The developer bypassed its limits – LsA magazine

Chip Apple The M4 hides a secret. The developer bypassed its limits – LsA magazine

https://www.letemsvetemapplem.eu/en/2026/06/19/mac-m4/

Publish Date: 2026-06-19 05:02:00

Source Domain: www.letemsvetemapplem.eu

Chips Apple Silicon are among the most powerful mobile processors in the world. But some of their capabilities remain hidden from developers behind official interfaces Applu. But now it turns out that the Neural Engine in the chip Apple M4 can be much more interesting than Apple commonly allows exploitation. A developer and security expert who goes by the nickname @0x0SojalSec claims that he has managed to bypass the chip’s common software limitations by reverse engineering it. Apple M4 and access the Neural Engine at a much lower level than standard tools allow Applu. The result should be the possibility of using the AI ​​accelerator not only for running ready-made models, but also for more demanding calculations associated with their training.

The Neural Engine is powerful, but Apple keeps him in check

Apple When introducing the M4 chip, Apple said its Neural Engine could handle up to 38 trillion operations per second. The company called it its fastest AI accelerator yet, emphasizing its use for AI tasks right on the device. The Neural Engine is one of the reasons why Apple can process some AI tasks locally on an iPad or Mac without the need to send data to cloudBut in practice, the Neural Engine has a clearly defined role. Apple It is used primarily for inference, i.e., for running already trained models, via public interfaces. If an application needs to recognize an object in a photo, transcribe speech, edit an image, or run a machine learning model, Neural Engine is an ideal helper. However, for directly training models, Apple does not commonly offer developers a simple and open approach.

According to available information, the developer did not take the classic route via Core ML or Metal. Instead, he was supposed to create his own low-level layer that could communicate with the hardware directly. In other words: he did not expect what he would get. Apple allows through official tools, but tried to find out what the chip is really capable…

Source