The processor leaks clues while it works
Encryption can be mathematically sound and still be implemented in a way that reveals information. During decryption, a CPU performs many small operations. Their timing and electrical activity can vary with the data being processed—and, in some implementations, with secret key bits. Those variations may produce measurable changes in power use, heat, or sound.
The sound is usually not the processor “saying” the key. Tiny shifts in electrical load can make components vibrate or alter other noise from a device. A recording may capture patterns too subtle for a person to hear; signal analysis can then look for repetitions linked to the processor’s activity.
From physical pattern to secret
An attacker needs to collect enough measurements while the target performs the relevant operation. By comparing those traces, the attacker may distinguish between different patterns of computation and use that evidence to narrow down the key. This is a side-channel attack: it exploits information emitted by the act of computing, rather than defeating the encryption mathematics directly.
The method is not a universal microphone trick. Leakage depends on the device, software, surroundings, and how often the attacker can observe it. Background noise can obscure a signal, and a program designed to take similar amounts of time and follow similar execution paths regardless of secret data can reduce some forms of leakage. Physical separation, noise, and other protections can also make measurement harder.
The important consequence is that encryption protects information only as well as the system handling it. A secret can remain hidden in the code and still leave a physical trace as the machine uses it.
