I've read almost everywhere that AES-256 can be used for Top Secret material (in the US). Is it really used or is it some kind of decoy to hide the more advanced algorithm they might use ?
Decoy for whom?
Security, and cryptography specifically, has a need for public scrutiny. It's been proven time and time again that hiding the nature of the protocol/algorithm/scheme doesn't provide any tangible security. Kerckhoffs' principle explains this: A cryptosystem should be secure even if everything about the system, except the key, is public knowledge.
It's "secure" up until someone finds out how it works and breaks it. Publically reviewed algorithms are something that people are always trying to break. The more people that try (and fail) to break something, the better that something is. This is important because it's really difficult to design crypto properly.
To answer your question specifically: Yes, AES is used for classified materials. Other algorithms are also used depending on the nature and use of the data.
A quick Google search came back with this PDF doc from NIST: Guideline for Implementing Cryptography In the Federal Government. It may be out of date, or there may be better documents, but it's an enlightening read.
Your question might be missing one plausible reason for governments to keep certain aspects of their cryptographic systems secret (besides the key), namely that their interest in learning the secrets of the opposition, might be as strong as their interest in keeping their own secrets to themselves. Hence, there are two questions that should be asked:
- Would revealing the block cipher algorithm you use as a building block in your schemes, potentially compromise your own security?
- Would revealing the block cipher algorithm you use as a building block in your schemes, potentially decrease your chances of compromising the security of the opposition?
The answer to the first question is no, for the reason SteveS wrote in his answer above. The second question is more interesting, if you are in to conspiracy theories. However, having seen a fair share of schemes implemented in real life application, I would not personally consider this aspect to be a concern. There are more ways AES has been used incorrectly that resulted in practically no security at all, than ways it has been used in a properly designed scheme that resulted in 256 bit security even after all design details, implementation details, deployment details and key management details had been take into account.