I think the major issues with this approach are: 1) You aren't sending them naked DNA which can be modified in a test tube, you're sending them epithelial cell

I think the major issues with this approach are: 1) You aren't sending them naked DNA which can be modified in a test tube, you're sending them epithelial cells that they extract the DNA from. 2) Any changes you make would mostly not be complete, and they're generating your reference genome from a consensus (to account for sequencing errors and DNA damage). This increases the odds that any randomization you do is basically swamped by the consensus algorithm. If the changes are not swamped then your decryption depends on knowing exactly which changes are from your encryption algorithm, and exactly which are from their consensus algorithm. 3) Bisulfite conversion, methyltransferases, general deamination, and the like make known changes to your genome that can easily be fully accounted for during analysis. There's no easily done "random" cryptographic technology that you can do on your DNA. And even if there was, it probably wouldn't be doable without already knowing your specific sequence.