I've proposed a syntax to support this feature in bug #65108.

See bug #63074 for a more general discussion.

  The bug is in man (man-db) not groff!

Bjarni reported this issue to Debian about 4 years ago.

https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=906091

"8 bit characters from the '.tm' request are displayed as \[u00XY]"

In bug #62726, I proposed a means of dealing with this.

> Having a stronger `troff` request to sanitize any diversion, string, or macro of node data is a feature that I am increasingly coming to think is both feasible and desirable.  Think, instead of "asciify", "utf8ify".  Anything that can be back-converted to ASCII or a UTF-8 sequence (not forgetting that our old friends the hyphen-minus, caret, tilde, etc. are not ASCII unless remapped) is, and everything else is thrown out.

> I relatedly think that arguments to the `tm` family of requests (including `ab`) should be similarly handled.  I think it would be significant effort for little benefit to add general localization support for troff output to the standard error stream.  That is, I don't want troff to have to care whether the environment uses Latin-1 or Latin-9 or Unicode.  So the argument(s) to these requests would be scooped into a temporary anonymous troff string and then "utf8-sanitized" as described above, then re-emitted.   As a first cut I wouldn't even inspect the environment, but just blast out the bytes in UTF-8 and if somebody's terminal encoding ain't that, they get mojibake.