As always, the obsession with firsts is a fools errand. The IBM 801 and the CDC 6600 are often mentioned as some of the earliest candidates. As so often with technology, it's a function of the context. In the case of RISC, integrated circuits were reaching capabilities that made it worthwhile reconsidering the conventional wisdom.

The 6600 didn't have general purpose registers. It had operand, address and increment registers. The 801 development began in October 1975. The Cray 1 had been in development for 4 years when it was released in 1975.

So I'll go with the Cray as being to OG RISC. But yes, firsts don't really matter except to say it sure wasn't the Berkeley RISC or the Stanford MIPS.

Never seek after "firsts" too avidly. I've seen people claim the rock is the first computer, and an IBM mainframe running VM/CMS is the first personal computer. (You get your own personal CMS instance running on the massive, institutional big iron, see?)