Various items that work with text allow the user to specify the encoding, so that agent could convert it to UTF-8 first to uniformly process this text line-by-line.
If user makes an error and specifies an item with non-existing encoding like:

vfs.file.contents[/home/zabbix/files_encodings/hosts,BADGER_3923]

then iconv library will detect that this encoding is invalid, but Zabbix will ignore this error and simply return the text as it was already converted:

zbx_convert_to_utf8()
...
        if ('\0' == *encoding || (iconv_t)-1 == (cd = iconv_open(to_code, encoding)))
        {
                res = cd;
                memcpy(out, in, in_size);
                out[in_size] = '\0';
                return out;
        }

Same is for agent 2.

This would result zabbix treating lets say UTF-16 text as UTF-8.
Of course this would lead to various undesired behaviours.

Zabbix agents must not ignore the iconv_open() function -1 error code and make an item unsupported.

Issue was raised in ZBX-21547 , check subissues 1 and 4