The Unicode range uFB50 to uFDFF is the Arabic Presentation Form A block.
UFCB0 to UFCBF displays as "ﲰﲱﲲﲳﲴﲵﲶﲷﲸﲹﲺﲻﲼﲽﲾﲿ" (remember arabic writes rigth to left). There are a number of unassigned codepoints in this block, namely : uFD40 to uFD4F, uFD90 to uFD91, uFDC8 to uFDEF, uFDFE to uFDFF. All of the assigned codepoints in the block range are arabic ligatures having category Lo (letter other), except uFDFC which is the RIAL SIGN currency symbol, category Sc (Symbol currency).
No codepoint should compare to the empty string: a codepoint is a codepoint and can't never be equal to "no codepoint", ever. That a codepoint is currently unassigned in today's Unicode version doesn't preclude it being meaningful in the future. And valid codepoints in a string have no right to disappear nor hide themselves from scrutiny.
But there is a catch here: AutoIt = comparison is case-insensitive. If you replace the test with the case-sensitive == you get a different, correct result.
I don't know how = is coded internally but yes, there is a bug here. Please file a ticket. Here's how to list failing (UCS2) codepoints:
For $i = 0 To 0xFFFF
If ChrW($i) = '' Then _
ConsoleWrite(ChrW($i) & @TAB & 'Chrw(0x' & Hex($i, 4) & ') = "" ? --> ' & (ChrW($i) = '') & @TAB & @LF)
Next