Most devices support privacy extensions (temporary v6 addresses) for this exact reason. https://www.rfc-editor.org/rfc/rfc4941
Behind a NAT, observers can only make that connection (using only addresses) for the network as a whole, instead of an individual device on the network. So the privacy extensions are weaker than NAT
(If the IPv6 privacy extensions used a different address for each connection, they would be more like NAT in this regard.)
That said, other observable clues still allow connections from a single device to be associated, NAT or not. There's TCP OS fingerprinting for example, and the close timing of related connections.
In this scenario it's a lot more work trying to map out your infrastructure than to try performing a MITM connection against some of your outbound traffic.