When/why did Lisps start using semicolons for comments?

Semicolons look to be statement sequencers (rather than terminators) in Algol68.

LISP 1.5 was punch card based, so comments probably would've been written on the cards themselves, I think? The manual gives no indication that there was a mechanism for in-language comments.

The Stanford Lisp 1.6 manual shows semicolon comments.

As for the origins, I'd look to see what if anything early assemblers used for indicating comments. Certainly the semicolon is common enough in current ones.

Maclisp from 1974 had semicolon as comments.

http://www.softwarepreservation.org/projects/LISP/MIT/Moon-MACLISP_Reference_Manual-Apr_08_1974.pdf