Intrinsic::ID enums are not encoded in the bitcode file, so I am not sure what why it would break. Ignoring the new target, a later release may add new intrinsics to existing targets and that also renumbers all intrinsic IDs and I assume that does not break the LTO flow. It seems to me that Intrinsic::ID enums are only required to be consistent within an LLVM build. Even the C API does not expose them directly, but you need to query the ID of an intrinsic using its name first (or an existing Function).