UTF-8 does not require BOM, but for UTF-16 and UTF-32 BOM is always present. Files with UTF-16 and UTF-32 without the BOM should be identified as binary.

But why there are no plans to support UTF-16 and UTF-32? Diff is part of the Git and is used all over the world. Now 2018 and Unicode solved problems with encodings.

Keepun wrote:

Files with encoding greater than 8 bits without BOM at the beginning can be immediately identified as binary.

No, the BOM is not required or recommended in UTF-8, so it would be a mistake to identify GNU/Linux text files as binary merely because they lack a BOM. Typically these files do not have a BOM, and when they do one of the first things many users do is remove the BOM because it can cause trouble in practice.

Diffutils does not support UTF-16, where a BOM would make more sense, and there are no plans to add support for UTF-16 (or for UTF-32, for that matter).