Officially RFC 6598 says
Devices MUST be capable of performing address translation when
identical Shared Address Space ranges are used on two different
Most NAT implementations are not capable of handing that case at least not witout extra hacks (on linux for example I belive that to implement NAT with overlapping internal and external space you would need to NAT the traffic twice in two different network namespaces).
You are of course free to ignore that paragraph and use the addresses anyway. RFCs are not laws. Using "shared address space" for your internal networks is certainly a lesser evil than using squat space.
If you do choose to ignore it and your ISP changes your connection to CGN then there is the risk of an addressing conflicts.
So like many things it comes down to a risk assessment. How badly do you need extra private address space? how likely is it that your internet connection will be put behind a CGN in the future?
If you are thinking about using this block because you are short of regular private addresses it's probably time for a long hard look at your IP addressing policies. Do you really have millions of devices on your internal networks? Are you wasting IPs with oversized allocations? isn't it time you thought about IPv6?