You’re probably right, but steganography with FEC should be enough to do the job; any predictive text errors would be caught with the checksumming.

After all, Phil Zimmerman got the entirety of the PGP source code from the US to Germany as a book. OCR combined with predictive text reconstruction has come a LONG way since then. The big problem today with OCR is that it often corrects errors that were present in the original document.